the effect of uncertainty on the occurrence and spread of

THE EFFECT OF UNCERTAINTYON THE OCCURRENCE AND

SPREAD OF FINANCIAL CRISES

Inaugural-Dissertation

zur Erlangung des Grades

Doctor oeconomiae publicae (Dr. oec. publ.)

an der Ludwig-Maximilians-Universitat Munchen

2007

vorgelegt von

Friederike Kohler-Geib

Volkswirtschaftliche Fakultat

Referentin: Prof. Dr. Monika Schnitzer

Korreferent: Prof. Jaume Ventura, Ph.D.

Datum der mundlichen Prufung: 29. Januar 2008

Promotionsabschlussberatung: 06. Februar 2008

Acknowledgements

This research project would not have been possible without the kind support

of a number of people to whom I would like to express my gratitude. First

and foremost, I would like to thank my supervisor Monika Schnitzer for her

excellent guidance, very helpful advice, and continuous support. Just as much,

I am grateful to Jaume Ventura for hosting me twice as visiting Ph.D. student

at University Pompeu Fabra, Barcelona, in spring 2006 and 2007 and for having

joined in guiding, advising, and supporting my work.

Furthermore, I would especially like to thank Fernando Broner, Antonio Ci-

ccone, Jarko Fidrmuc, Harald Fadinger, Pablo Fleiss, Georg Gebhardt, Michael

Grimm, Frank Heinemann, Florian Heiss, Christian Holzner, Gerhard Illing,

Francisco Penaranda, Thijs van Reens, Stefan Schubert, Luis Serven, Andrei

Sleptchenko, Frederik Udina for fruitful discussions and very helpful comments

and suggestions. Then I am thankful to Jeromin Zettelmeyer for inspiring me

to start working on sudden stops of capital flows. Special thanks go to Robert

Holzmann for continuous, very helpful advice. My thanks extend to my col-

leagues at the University of Munich, the University Pompeu Fabra, the Munich

Graduate School of Economics, and at the Chair of Comparative Economics

at the University of Munich, who have been an important source of feedback

and encouragement.

Additionally I would like to thank participants at various research semi-

nars at the University of Munich and the University Pompeu Fabra, as well as

during sessions at the LACEA Annual Meeting 2005 in Paris, the 10th Interna-

tional Conference on Macroeconomic Analysis and International Finance 2006

in Rethymno, the 4th Workshop on Monetary and Financial Economics 2006

in Halle, the Annual Meeting of the Verein fur Socialpolitik 2006 in Bayreuth,

and the Spring Meeting of Young Economists 2007, who gave me valuable

comments.

Moreover, I would like to thank Gaston Gelos, Alexandro Izquierdo, Luis

Gonzalo Llosa, and Anna Stangl for sharing with me and giving information

on part of the data set used in the empirical parts of this research. In addition,

I am grateful to Marybeth Shea for excellent editing.

I am very thankful for the support and the provision of a great research

environment by the Munich Graduate School of Economics. I would like to

thank Ingeborg Buchmayr, Gabriella Szantone-Sturm, and Agnes Bierprigl for

help and assistance. Special thanks go to Dirk Rosing and Peter Dumitsch for

taking care of every major or minor computer problem.

I gratefully acknowledge the financial support by the German Research

Foundation (DFG), the German Academic Exchange Service (DAAD), the

European Commission, and the Kurt Fordan Foundation, which made this

research possible.

Finally, but most importantly, I am deeply grateful for the love, encour-

agement, and support of my family. With all my heart I would like to thank

my husband, Malte Geib, for the best company that I can imagine in a thesis

endeavor. I am also particularly thankful to my mother, Gertrud Kohler, to

whom I dedicate this work.

Washington DC, October 2007

Contents

1 Introduction and Literature Overview 1

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.1.2 Aims and Scope . . . . . . . . . . . . . . . . . . . . . . . 3

1.1.3 Outline of the Study . . . . . . . . . . . . . . . . . . . . 5

1.2 The Occurrence of Financial Crises . . . . . . . . . . . . . . . . 6

1.2.1 Crises Phenomena . . . . . . . . . . . . . . . . . . . . . 6

1.2.2 Generations of Crises Models . . . . . . . . . . . . . . . 9

1.2.3 Modeling Approaches to Sudden Stops . . . . . . . . . . 10

1.2.4 Empirical Analyses of Sudden Stops . . . . . . . . . . . . 14

1.2.5 Contribution of this Study . . . . . . . . . . . . . . . . . 19

1.3 The Spread of Financial Crises . . . . . . . . . . . . . . . . . . . 21

1.3.1 The Contagion Phenomenon . . . . . . . . . . . . . . . . 22

1.3.2 Modeling Approaches to Contagion . . . . . . . . . . . . 23

1.3.3 Contagion Channels . . . . . . . . . . . . . . . . . . . . 24

1.3.4 Contagion through Common Creditors . . . . . . . . . . 26


1.4 The Effect of Uncertainty . . . . . . . . . . . . . . . . . . . . . 30

1.4.1 The Concept of Uncertainty . . . . . . . . . . . . . . . . 31

1.4.2 Uncertainty, Crises and their Spread . . . . . . . . . . . 32


i

ii CONTENTS

2 Uncertainty and Occurrence of Crises 39

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

2.2 Theoretical Background . . . . . . . . . . . . . . . . . . . . . . 45

2.2.1 The Firms . . . . . . . . . . . . . . . . . . . . . . . . . . 45

2.2.2 The Government . . . . . . . . . . . . . . . . . . . . . . 47

2.2.3 The Reduced Form Game Between Firms . . . . . . . . . 47

2.3 The Common Knowledge Game . . . . . . . . . . . . . . . . . . 48

2.3.1 High Growth and Low Growth Equilibrium . . . . . . . . 48

2.3.2 The Tripartite Classification of Fundamentals . . . . . . 48

2.4 The Private Information Game . . . . . . . . . . . . . . . . . . 50

2.4.1 Informational Structure . . . . . . . . . . . . . . . . . . 50

2.4.2 Object of Optimization . . . . . . . . . . . . . . . . . . . 51

2.4.3 Unique Equilibrium . . . . . . . . . . . . . . . . . . . . . 52

2.5 Comparative Statics . . . . . . . . . . . . . . . . . . . . . . . . 55

2.5.1 Changes in the Technology Parameter α . . . . . . . . . 56

2.5.2 Changes in the International Interest Rate r . . . . . . . 57

2.5.3 Changes in the Degree of Uncertainty ε . . . . . . . . . . 60

2.6 Testable Hypotheses . . . . . . . . . . . . . . . . . . . . . . . . 61

2.7 Empirical Analysis . . . . . . . . . . . . . . . . . . . . . . . . . 61

2.7.1 The Data . . . . . . . . . . . . . . . . . . . . . . . . . . 62

2.7.2 Benchmark Regression . . . . . . . . . . . . . . . . . . . 65

2.7.3 Analysis with Yearly Data . . . . . . . . . . . . . . . . . 67

2.7.4 Analysis with Monthly Data . . . . . . . . . . . . . . . . 68

2.7.5 Facing Endogeneity . . . . . . . . . . . . . . . . . . . . . 69

2.7.6 Robustness Analysis . . . . . . . . . . . . . . . . . . . . 71

2.8 Policy Implications . . . . . . . . . . . . . . . . . . . . . . . . . 72

2.9 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73

2.10 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76

2.10.1 Value of the Firm . . . . . . . . . . . . . . . . . . . . . . 76

2.10.2 Lemma Proofs . . . . . . . . . . . . . . . . . . . . . . . . 77

CONTENTS iii

2.10.3 Iterated Elimination of Dominated Strategies . . . . . . 83

2.10.4 Figures and Tables . . . . . . . . . . . . . . . . . . . . . 85

3 Uncertainty and Spread of Crises 101

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101

3.2 The Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108

3.2.1 Model Setup . . . . . . . . . . . . . . . . . . . . . . . . . 110

3.2.2 Solving the Model . . . . . . . . . . . . . . . . . . . . . . 112

3.2.3 Results and Implications . . . . . . . . . . . . . . . . . . 115

3.2.4 Testable Hypotheses . . . . . . . . . . . . . . . . . . . . 117

3.3 Empirical Analysis . . . . . . . . . . . . . . . . . . . . . . . . . 117

3.3.1 The Data . . . . . . . . . . . . . . . . . . . . . . . . . . 117

3.3.2 Methodology . . . . . . . . . . . . . . . . . . . . . . . . 119

3.3.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123

3.4 Policy Implications . . . . . . . . . . . . . . . . . . . . . . . . . 132

3.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133

3.6 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135

3.6.1 Lemma Proofs . . . . . . . . . . . . . . . . . . . . . . . . 135

3.6.2 Iterated Elimination of Dominated Strategies . . . . . . 135

3.6.3 Figures and Tables . . . . . . . . . . . . . . . . . . . . . 137

Bibliography 149

iv CONTENTS

Chapter 1

Introduction and LiteratureOverview

1.1 Introduction

1.1.1 Motivation

Most major financial crises in the past three decades entailed devastating ef-

fects on real economic activities in affected countries and markets. Examples

are the Mexico crisis in 1994, the crises in South East Asia in 1997, and the

Turkey crisis in 2001.1

In particular, financial crises that comprise a sudden stop of capital flows,

i.e., a sharp negative variation in capital flows, are characterized by severe and

long lasting economic effects. For example, the Mexico crisis, characterized

by a sudden stop and a currency crisis, led to a fall of real equity prices in

CPI units (Consumer Price Index) of 29 percent, in industrial production of

10 percent, and to a plunge of 6.5 percent in private consumption.2 In 1995,

Mexico’s GDP declined 6 percent as compared to the previous year.3

1See Table 2.12 in Chapter 2 for an overview on the most severe financial crises of the lastthree decades and their effects on the real economic activity in the crises countries. Thesecrises were so severe that they appeared in newspaper headlines around the world and areremembered for the accompanying turmoil. Kaminsky, Reinhart, and Vegh (2000) surveysthose crises that seized several economies.

2Mendoza and Smith (2006) report these values comparing late January 1995 to April1994.

3See Table 2.12 in Chapter 2.

1

2 CHAPTER 1. INTRODUCTION AND LITERATURE OVERVIEW

Many financial crises have occurred in emerging market economies around

a few initial crises, particularly around the Mexican crisis in 1994, the Thai

crisis in 1997, and the Russian crisis in 1998, as well as in developed economies

around the breakdown of the European Exchange Rate Mechanism in 1992.

These periods of crises concentration suggest contagion effects, i.e., the trans-

mission of crises across countries beyond what would be implied by common

shocks.4

The high economic cost of these periods explains the effort in trying to

understand the factors behind the occurrence and the spread of crises. Many

researchers have analyzed factors explaining the occurrence of financial crises

and their effects on the real economy, and have thought about possible poli-

cies that may prevent crises or mitigate their effects. Just as much effort

has been exerted on exploring corresponding questions regarding the spread

of crises. The present study seeks to contribute to this effort by analyzing

one specific factor neglected so far but potentially delivering fresh insights on

policies preventing the occurrence and the spread of crises: uncertainty about

the fundamentals. In the present study, uncertainty about the fundamentals or

uncertainty refers to the disagreement of private investors about the state of

the fundamentals of an economy.5

Uncertainty about the fundamentals between private investors belongs to

the variables only recently discovered as potential explanatory factors of cur-

rency, debt, and banking crises. Its role in the transmission of crises across

countries has even been entirely neglected so far. Financial crises occurring

earlier, such as currency crises of Mexico and Argentina in the 1970s, could be

largely explained by inconsistent economic policies or bad fundamentals in the

crises countries. Later, crises such as the breakdown of the European Exchange

Rate Mechanism in 1992 appear to have been triggered by self-fulfilling beliefs

of speculators or investors. However, the Mexican crisis in 1994, and to an

even larger extent the Asian crisis in 1997, have shown that previous models

were not sufficient to explain all crisis features. In particular, the sudden stop

of capital flowing to affected countries and the spread of a financial crisis from

one market to another could not be explained by previous models. Therefore,

the search for crisis triggers has been extended to factors within international

capital markets and investor behavior, uncertainty being one of these.

4See Didier, Mauro, and Schmukler (2006) for this definition.5This definition is widely used in the literature of global games.

1.1. INTRODUCTION 3

Studies modeling currency crises, debt crises, and banking crises as coor-

dination games have contributed promising approaches to the prevention of

these crises phenomena.6 This modeling approach allows for an analysis of

the effect of uncertainty. Sudden stops and contagion have not been analyzed

from this angle.

Decisions underlying the occurrence of sudden stops of capital flows are

similar to those underlying currency, debt, or banking crises.7 This renders

the analysis of uncertainty in the context of sudden stops possible. However,

factors driving sudden stops and the crisis phenomenon itself differ profoundly

from other crises phenomena. Therefore, analyzing the phenomenon of sudden

stops in terms of a coordination game between private agents promises to

generate new, insightful results.

While the spread of financial crises has been analyzed in the setup of a co-

ordination game, the literature stops short of analyzing which role uncertainty

plays in the transmission.8 In addition, the spread of stock market drops has

not yet been modeled in such a setting. However, such an analysis appears

promising. In particular, it can help in understanding the spread of crises

across countries that are unrelated in terms of their fundamentals.

Furthermore, the empirical quantification of the effect of uncertainty on the

occurrence of sudden stops and on the spread of crises is missing in previous

research. The goal of the empirical parts in chapter 2 on sudden stops and of

chapter 3 on the spread of stock market crises address this lack.

1.1.2 Aims and Scope

The overall objective of this study is to explore the effect of one particular

possible explanatory factor of financial crises and their spread: uncertainty

about the fundamentals.

6See section 1.4 for the concrete policy implications of coordination games analyzing theeffect of uncertainty such as by Heinemann and Illing (2002) in the context of currency crisis,or by Morris and Shin (2004) on debt crises.

7Concretely, the decision of abstaining from investment in a particular country’s assetspositively depends on other agents choosing the same strategy and negatively depends on thequality of the fundamentals in the respective economy. These features similarly characterizethe choice of attacking a currency, of abstaining from rolling-over debt contracts, or ofwithdrawing bank deposits.

8See Goldstein and Pauzner (2004) for a prominent example.


Achieving this overall objective can be split into a number of aims with

smaller scope.

A first aim is to help in understanding how sudden stops of capital flows

are triggered. In particular, I want to show how coordination failure between

private investors can contribute to the occurrence of sudden stops. I focus on

the effect of uncertainty about the fundamentals.

The second aim is then to understand the relevance of uncertainty in ex-

plaining sudden stops. The large number of factors potentially functioning

as crisis triggers requires a rigorous empirical analysis, taking into account a

large number of control variables. Moreover, the potential problem of reverse

causality running from a sudden stop towards uncertainty requires a careful

robustness analysis.

The third aim then is to provide policy implications based on a new under-

standing of the occurrence of sudden stops and their relevance. The ultimate

goal of these policy implications is to contribute to the prevention of sudden

stops and to the mitigation of their consequences.

The fourth aim is to illustrate the mechanism through which uncertainty

about fundamentals propagates financial crises across markets. In this context,

I focus on contagion of stock market crises. These crises, together with banking

crises, appear to be at the core of spreading crises.9

The fifth aim of this study is to validate empirically the effect of uncer-

tainty on the spread of stock market crises. Again, the existing alternative

factors of contagion and the risk of reverse causality call for rigorous empirical

verification and careful robustness analysis.

The sixth aim is, again, to provide policy implications in the effort to

prevent contagion and mitigate its consequences.

The following two points are relevant for clarifying the scope of the present

study. Firstly, the present study does not claim that uncertainty about the

fundamentals is the only factor either triggering or transmitting financial crises.

Rather, the purpose of my study is to explore the effect of one factor neglected

so far. As the empirical analysis here shows in both cases, in the occurrence of

sudden stops of capital flows and the spread of stock market crises, the effect

is not negligible in terms of magnitude. Additionally, analysis of uncertainty

9Portfolio flows and bank lending have been found to be particularly diversified acrosscountries and volatile during crises periods. For evidence, see Levchenko and Mauro (2006).

1.1. INTRODUCTION 5

about the fundamentals can help identify additional policy ideas that diminish

the probability of a crisis or its propagation and mitigate the consequences.

Secondly, the models in this study will not explain all crisis features char-

acterizing the recent headline crises. In particular, the models do not accom-

modate bank runs or debt crises. Both bank-run crises and roll-over debt

crises have already been analyzed in light of coordination failure. This study

focuses on phenomena shown to be especially crucial to the financial turmoil

of the recent headline crises and which have the potential to generate new

insights. This narrowed, strategic focus allows for concrete results and for

drawing concrete policy implications.

1.1.3 Outline of the Study

This study consists of three chapters. The rest of chapter 1 briefly consid-

ers relevant threads of literature. Reviewed first is the relevant literature on

financial crises. Then follows a summary of the literature on the spread of

financial crises. Finally, the first chapter concludes with an overview of the

existing literature on the effect of uncertainty on the occurrence and the spread

of financial crises.

Chapter 2 addresses the effect of uncertainty about the fundamentals on

the probability of sudden stops of capital flows, both from a theoretical and

an empirical perspective. A coordination game with private information about

the fundamentals is set up to show the effect of investment safety, of the inter-

national interest rate, and of uncertainty on the probability of a sudden stop

of capital flows. The model predicts that the probability of a crisis decreases

with an increase in investment safety. However, the crisis probability increases

with an increase in the international interest rate. Moreover, the crisis prob-

ability increases with an increase in the uncertainty about the fundamentals,

i.e., with the dispersion of private signals about the true value of the funda-

mentals. The analysis uses two data sets of Consensus Economics and WES

(World Economic Survey) forecasts for 31 developed and developing coun-

tries from January 1990 to December 2001 to test the theoretical predictions.

Applying probit estimations controlling for time and country effects makes

the validation of the theoretical predictions possible. Additionally, results are

tested for robustness across many specifications.

Chapter 3 analyzes the impact of uncertainty on the spread of stock mar-


ket crises, both theoretically and empirically. The effect of uncertainty about

the fundamentals on investment decisions is an important cause of financial

crises propagating across countries. Firstly, a coordination game on invest-

ment illustrates the increasing effect of a surprise crisis in one country on the

probability of a crisis in a second country through higher uncertainty there. An

anticipated initial crisis generates the opposite effect. Secondly, these theoret-

ical predictions are tested empirically. Fixed effects panel estimations validate

the impact of the initial crisis on uncertainty in potentially-affected countries.

Subsequently, probit estimations confirm the positive impact of uncertainty on

the crisis probability in the affected economy. The results are robust across

various specifications.

1.2 The Occurrence of Financial Crises

The literature on financial crises is extensive. Therefore, this overview men-

tions only the most important studies shaping the understanding of these

events. A particular emphasis lies on the literature on sudden stops of capital

flows as they are at the core of interest in chapter 2 of this study. This section

only briefly revisits the literature on stock market crises. As the spread of

stock market crises is at the core of interest in chapter 3 of this study, readers

will find more detail on this literature in the literature overview on contagion

in section 1.3. In addition to reviewing relevant literature, this section puts

the present study in perspective by defining its scope in the context of the

literature.

This section is organized as follows: First, the most important types of

crises are defined. Second, the different steps in the development of crises

models are traced. Third, this section treats the literature on sudden stops of

capital flows, dividing the presentation into theoretical and empirical studies.

1.2.1 Crises Phenomena

In the literature, currency crises that have long been considered at the core

of larger financial turmoil cases are defined as substantial exchange rate de-

valuations. In the empirical analyses of currency crises, researchers have mea-

sured the devaluations by so-called exchange market pressure indices. A crisis

is defined as a significant change in the index. While a few authors consider

1.2. THE OCCURRENCE OF FINANCIAL CRISES 7

changes in the nominal exchange rate exclusively10, most authors, additionally,

take into account either changes in foreign exchange reserves11 or changes in

foreign exchange reserves and in the interest rate12 to accommodate possible

policy responses to initial pressure on the nominal exchange rate.

Banking crises or panics have been accurately defined as events in which

”bank debt holders suddenly demand that banks convert their debt claims into

cash to an extent that the banks are forced to suspend the convertibility of

their debt into cash” by Calomiris and Gorton (1991). Caprio and Klingebiel

(2002) define systemic banking crises as ”much or all of bank capital being

exhausted,” which is a definition more easily measured.

Most authors understand a debt crisis as a credit event defined as a non-

repayment of pre-agreed debt service.13 Pescatori and Sy (2004) give an exten-

sive overview of the definitions of debt crises, categorizing the corresponding

credit events as sovereign defaults, large arrears, large International Monetary

Fund loans, and as distress.

Kaminsky and Reinhart (1999) introduced the term twin crisis to the

literature. Twin crisis describes the simultaneous occurrence of a banking and

a currency crisis. In their empirical analysis Kaminsky and Reinhart (1999)

find that the banking crisis in most cases precedes the currency crisis. This

observed order of events is often used to explain why the focus of a part of the

literature has shifted from currency to banking crisis as the core of financial

turmoil. A more recent literature also analyzes the simultaneous occurrence

of debt and currency crises.14

The term sudden stop was first introduced by Dornbusch, Goldfajn, and

Valdes (1995). Most authors define a sudden stop of capital flows as a sharp

reversal in capital flows associated with severe economic consequences. How-

ever, authors differ in their sudden stop definitions regarding the distinction

between crisis features and the consequences of the crisis. While, for example,

Calvo (2003) defines a sudden stop simply as a large reduction in the flow of

international capital considering the ensuing turmoil as a crisis consequence,

Mendoza and Smith (2006) include three stylized facts into their definition

10See Frankel and Rose (1996).11See, for example, Berg and Patillo (1998).12See, for example, Eichengreen, Rose, and Wyplosz (1996) or Fratzscher (2003).13See Sachs (1984) for this definition.14See, for example, Reinhart (2002) or Bauer, Herz, and Karb (2007).


of a sudden stop: a sudden, sharp reversal in capital inflows and the current

account, large declines in absorption and production, and collapses in both

real asset prices and in the price of non-tradable goods relative to tradables.

Henceforth, sudden stop in this study means the reversal in capital flows. The

study will treat the consequences separately.15

Sudden contractions in current account deficits or even current account

reversals are closely linked to sudden stops in capital flows. Capital inflows

equal the current account deficit plus the accumulation of international reserves

by national accounting, if abstracting from errors and omissions. A sudden

stop in capital inflows must be met by a lower current account deficit or by

reserve losses. As Calvo and Reinhart (2000) illustrate, both cases happen

in reality. Edwards (2004b) demonstrates that current account reversals and

sudden stops in capital inflows are highly correlated. A part of the empirical

literature uses reversals in the current account to identify sudden stops of

capital flows.16

The loss in foreign reserves is defined as a balance of payment crisis

if it is severe: Krugman and Obstfeld (2003) define it as a sharp change in

official foreign reserves sparked by a change in expectations about the future

exchange rate.17 Clearly, then, a balance of payment crisis is also closely linked

to a currency crisis that occurs when depletion of reserves is not sufficient to

buffer the shock: the currency, then, depreciates.

Practitioners define stock market crises as precipitous drops in market

prices or economic conditions.18 An academic definition is more difficult as

Mishkin and White (2003) put it: ”On the face of it, defining a stock market

collapse is simple: when you see it, you know it. However, a precise definition

is more difficult.” In this study, following Broner, Gelos, and Reinhart (2006),

a stock market crisis is defined as a significant drop in stock market returns.

15The empirical analysis in the second chapter employs two alternative measures of suddenstops. The first measure is based on the reversal in capital flows exclusively. The secondmeasure, additionally takes negative GDP growth into account.

16See for example Hutchison and Noy (2006).17See Krugman and Obstfeld (2003), p. 502.18See http : //www.investorwords.com for this definition.


1.2.2 Generations of Crises Models

The search for models explaining the occurrence of financial crises has evolved

in several waves or generations. These generations follow the occurrence of

financial crises not explained by previous models. The intellectual development

is stepwise, however, and relies on prior modeling work.

In the first generation models, economic fundamentals and unsustainable

domestic policies are at the core of financial crises.19 These models respond to

the currency crises of Mexico and Argentina in the 1970s. In these models, the

combination of government deficits financed with the help of seignorage and

fixed exchange rate regimes leads to a depletion of finite foreign reserves. Ex-

ternal investors generate a speculative attack when reserves fall below a critical

level to avoid capital losses in case of the inevitable collapse. This results in a

currency crisis.

In the second generation models, crises occur driven by self-fulfilling be-

liefs. The models that were developed after the severe speculative attacks on

the European Exchange Rate Mechanism in 1992 take into account that the

attacked economies are mainly characterized by government surpluses and sub-

stantial foreign reserves. In Obstfeld (1994), for example, crises result from the

tradeoff between a fixed exchange rate and a desire for an expansionary mon-

etary policy to lower unemployment. The crisis, then, stems from investors

suspecting that the cost of defending the peg for unemployment becomes too

high for the government and that the peg may have to be abandoned. The

resulting pressure on interest rates can be sufficient to force the government

into devaluation.

In the third generation models, fragility of financial structure and insti-

tutions is at the core of the crisis. Most approaches in this generation combine

features of the first and the second generation in the sense that models are

fundamentals- and belief-driven at the same time. In particular, the Asian

crisis in 1997-98 cannot be explained by first or second generation models be-

cause, prior to the crisis, the affected economies are seemingly characterized

by sound fundamentals and, unlike the European countries prior to the 1992

crisis, by low rates of unemployment. Additionally, the 1994 Mexican as well

as the 1997-98 Asian crises both involve sudden stops of capital flows and con-

tagion. This is why the literature on financial crises broadened widely. As far

19See Krugmann (1979) and Flood and Garber (1984).


as crises models are concerned, the papers reviewed in the following literature

overview are all part of the third generation.

1.2.3 Modeling Approaches to Sudden Stops

The theoretical literature on sudden stops of capital can be distinguished into

two broad threads. One thread features dynamic, stochastic, general equilib-

rium (DSGE) models, which can only be solved numerically. In these models,

sudden stops originate in the interaction of productivity shocks with financial

market frictions such as credit restrictions and transaction costs of assets.20

Another thread on sudden stops features crisis models based on multiple

equilibria, which are solvable analytically. In these models, the occurrence of

sudden stops stems from domestic vulnerabilities interacting with exogenous

shocks, such as terms of trade shocks or an increase in country risk.21

Dynamic, Stochastic, General Equilibrium Models

The first large thread of sudden stop literature consists of DSGE models where

the sudden stop occurs as an equilibrium outcome if financial constraints bind

and Fisher’s debt-deflation mechanism is triggered as a result.22

If the economy is highly leveraged, i.e., the ratio of debt and/or working

capital to asset values is sufficiently high, an adverse shock, i.e., a negative

productivity shock, of standard magnitude, with financial constraints being

binding, may lead agents to liquidate capital to fulfill margin calls.23 The

20See, for example, Mendoza and Smith (2006). Mendoza (2006a) delivers a survey onthis literature.

21The most important papers in this literature are Calvo (1998a), Calvo (2003), Calvo,Izquierdo, and Mejia (2004), Calvo, Izquierdo, and Talvi (2006), and Caballero and Krish-namurthy (2001). While the first four papers are all based on balance sheets consideration,Caballero and Krishnamurthy (2001) model a liquidity crisis from the perspective of privateinvestors within an economy subject to domestic and foreign collateral constraints. The roleof collateral constraints in the sudden models is reviewed for DSGE models. Hence, theCaballero and Krishnamurthy (2001) paper is not reviewed further here.

22DSGE models are a further development of small open economy business cycle modelswhere the most prominent examples are Chari, Kehoe, and McGrattan (2005), Christiano,Gust, and Roldos (2004), and Neymeyer and Perri (2004). In these models, collateralconstraints are also at the core of financial crises. The DSGE literature is the more recentone, deserving concentration here.

23For example, in Mendoza and Smith (2006) these constraints limit external debt toa fraction of the market value of domestic equity holdings. In Mendoza (2006b) a first


sudden high supply of assets reduces the price of capital, thereby tightening

the constraint. This state sets off a spiraling collapse of investment and as-

set prices, which has strong real effects: The current account and domestic

absorption immediately reverse. Future levels of capital, output, and factor

demands fall due to the initial investment decline. In addition, the collapse

of the value of collateral assets can render a collateral constraint on working

capital binding, thereby inducing a decline in production and factor demands.

These types of models suggest policies aimed at preventing deflation as a

good option. For example, Durdu and Mendoza (2006) study price guarantees

on the emerging asset class that stop the debt-deflation process by introduc-

ing a moral hazard-like distortion affecting foreign traders. They find that

welfare-improving guarantees require complex state-contingent features. The

indexation of debt to GDP or to commodity prices and hard currency adoption

have also been suggested as remedies against sudden stops. Another recom-

mendation, stemming from debt-deflation models, is the build-up of foreign

reserves to minimize the long-run probability of sudden stops of capital flows.

However, as pointed out, for example, by Caballero and Panageas (2006), this

prevention strategy is a very expensive insurance against sudden stops.

One major drawback of these types of models is that they do not deliver

close form solutions. They can only be solved numerically and, therefore,

do not provide strong results. Additionally, introducing private information

about the fundamentals into these models is difficult. Thus, an analysis of

the effect of uncertainty about the fundamentals in such a setup would be

highly complicated, without the strong likelihood of producing more convincing

results. Therefore, this research builds on the second thread of the literature,

models based on multiple equilibria.

Models Based on Multiple Equilibria

The second large thread of literature consists of models where the occurrence of

a sudden stop is based on multiple equilibria. A crisis materializes in the event

of a discontinuous switch between equilibria. The mechanism through which

the sudden stop occurs in these models is the following: Before the crisis, the

economy displays a current account deficit. A growth collapse induced by an

collateral constraint limits debt not to exceed a fraction of the liquidation value of collateralassets; a second collateral constraint limits working capital financing to not exceed a fractionof the firms’ assets.


exogenous shock displays sudden stop features:24 The current account deficit

discontinuously switches to zero because, in the non-monetary version of these

models, the current account deficit equals the amount of capital inflows. As

nontradable goods are normal goods, demand for them declines as net wealth

declines due to the growth collapse. This implies that less tradable inputs are

used for the production of nontradables. This, in turn, increases the marginal

productivity of tradables such that their price increases relative to the price of

nontradables, i.e., the sudden stop will be accompanied by a real depreciation.

In the monetary version of these models, in which the current account deficit

equals capital inflows minus the accumulation of international reserves assets,

the slowdown of capital inflows associated with current account deficit could be

cushioned by a drop in reserves. If the latter is high enough, this is considered

a balance of payment crisis. However, in reality the reserves often do not suffice

to buffer the shock. This implies that a currency crisis would ensue.

The literature on sudden stops identifies domestic vulnerabilities as well as

systemic capital market forces as core explanatory factors behind a crisis. The

key domestic variables are an unsustainable fiscal policy (i.e., fiscal policies

that depend on high taxes, which make after-tax revenues unattractive to in-

vestors), short term foreign debt, contingent debt (i.e., government guarantees

for private loans that only become apparent in bad states of nature), liability

dollarization, and a high leverage of the current account deficit (i.e., low out-

put of tradables relative to a high demand for them within the country, which

makes a country vulnerable to real exchange rate fluctuations). The relevant

market forces have been detected as TOT (terms of trade) shocks, fluctuations

in the world market interest rates, or loss of confidence in emerging market

economies as a whole. Calvo, Izquierdo, and Talvi (2005) and Calvo et al.

(2006) measure the loss in confidence in emerging market economies by a sig-

nificant increase in the aggregate EMBI (Emerging Market Bond Index by JP

Morgan) spread and call it substantial turmoil in global capital markets.

The policy implications from this type of model follow from the detected

vulnerabilities that economies or their governments can try to avoid. In partic-

ular, foreign-denominated short term debt is a variable that emerging market

governments can influence and should reduce.25 Governments may find them-

24In Calvo (2003) the exogenous shock, i.e., a TOT (terms of trade) shock, drives theeconomy beyond a critical level of debt that is inconsistent with a high growth equilibrium.

25The discussion of foreign denomination of emerging market debt has been labeled orig-inal sin. Various articles regarding this topic can be found in Eichengreen and Hausmann


selves facing two major difficulties: first, the vulnerability against external

factors and, second, the difficulty in borrowing in their own currency. These

governments should consider the following two options: In the effort of mitigat-

ing crises effects or of hedging against the probability of crises, governments can

accumulate foreign reserves or they can use alternative financial instruments.

A recent discussion of reserves accumulation as policy tool in the context

of sudden stops has been subject to a series of papers. Caballero and Cowan

(2006) illustrate that self-insurance strategies such as reserves accumulation,

public de-leveraging and export promotion are inefficient external insurance

mechanisms. The authors suggest financial hedging instead. Caballero and

Panageas (2005) analyze reserves accumulation and hedging mechanisms to

mitigate the consequences of a sudden stop, whereas Caballero and Panageas

(2006) analyze the role of hedging in diminishing the probability of a crisis.

An older policy discussion emerging from these models deals with capital

controls and exchange rate management. Montiel and Reinhart (1999) discuss

advantages and disadvantages of capital inflow controls. However, the authors

find that the effects on the magnitude of flows are restricted. Calvo (2003)

discusses the incentive of a central bank to peg its exchange rate once a crisis

occurs to buffer a part of the current account adjustment. The central bank

might provoke a balance of payment crisis to be able to transfer reserves to

private agents of the economy.

Although these models generate numerous valuable insights, they are char-

acterized by a number of drawbacks and shortcomings. A first drawback of

this literature is that these models stop short of a framework to predict sudden

stops. Although these models discuss early warning indicators and determine

key vulnerabilities, they do not produce a theory that can attach a probability

of a crisis to these factors. This results from the models displaying multiple

equilibria without a theory-based selection procedure.

A second drawback concerns the lack of analysis of investor coordination.

The introduction of a representative agent eludes analysis. However, espe-

cially in the context of financial crises, what other investors in the market

do is paramount. Diamond and Dybvig (1983) have modeled bank runs as

a coordination game. This prominent and important analysis has generated

valuable insights into the occurrence of such crises and possible prevention.

(2005).


The decision to invest or not to invest in a particular country is characterized

by strategic complementarity in the same way as the decision to keep or with-

draw bank deposits. Therefore, analyzing issues of coordination and its failure

in the context of sudden stops of capital flows is a promising research agenda.

A third drawback along the lines of the second criticism is that the current

literature on sudden stops lacks analysis of the effect of uncertainty about the

fundamentals. Based on the prominent analysis of Morris and Shin (1998) on

the occurrence of speculative attacks, Heinemann and Illing (2002) show that

speculative attacks become more probable with a higher dispersion of private

signals about the true value of the fundamentals, i.e., the dispersion of beliefs

of investors on the status of the economy. Again, the decision to invest within a

country is similar to the decision to attack a currency. An analysis of the effect

of uncertainty in the context of a sudden stop of capital flows thus appears

promising.

1.2.4 Empirical Analyses of Sudden Stops

In the context of sudden stops of capital flows, two threads of literature are

relevant: In one thread stands the literature on factors that drive capital flows

in and out of emerging economies; in the second thread stands the literature

that analyzes the drivers and consequences of pronounced crises events.

Empirical Analyses of Capital Flows

Lopez-Mejia (1999) provides a concise survey on the magnitude, regional des-

tination, reversibility, and composition of large capital flows in the 1990s. The

paper shows the heavy concentration of capital flows to China, Brazil, Mexico,

Thailand, and Indonesia in 1990-97 and the similarities of their reversibility

between the 70s and the 90s. The paper then surveys the findings of the lit-

erature on pull and push factors of capital flows into emerging markets from

developed countries and vice-versa.

The literature on pull and push factors is concerned with the relative im-

portance of domestic versus external factors in explaining capital inflows to

emerging economies. The underlying concern is that large surges in inflows

make countries more vulnerable to financial crises in the case that capital

flows are driven primarily by factors outside the emerging markets. One of the


first papers of this literature is Calvo, Leiderman, and Reinhart (1993). In this

paper, the authors examine empirical evidence for 10 Latin American coun-

tries, finding that, for the analyzed period, foreign factors, in particular the

US interest rate and GDP growth, account for 30 to 60 percent of the variance

in real exchange rates and reserves (two variables that directly reflect develop-

ments in the financial account) depending on the country. Chuhan, Claessen,

and Mamingi (1987) find a stronger effect of domestic variables. They include

Latin American and Asian countries into their analysis. They find that do-

mestic factors in their analysis such as country credit rating, secondary bond

prices, and price earning ratios in the domestic stock markets explain about

half of the bond and equity flows from the United States to a panel of six

Latin American countries. For Asia, they conclude that domestic factors ac-

count for about two thirds of bond and equity flows into the region. Building

on this, Fernandez-Arias (1996) decomposes the improvements in creditwor-

thiness found in Chuhan et al. (1987) into those stemming from a decline in

global interest rates and those arising from improvements in the domestic en-

vironment. He finds that the interest rate accounts for around 86 percent of

the increase in portfolio flows for the average emerging market during 1989-93.

In particular, the literature addresses the determinants of foreign direct

investment (FDI).26 The special interest in FDI originates from this quality:

FDI is the least volatile form of capital flows as compared to portfolio equity

investment, portfolio debt investment, other flows to the official sector, other

flows to banks, and other flows to the non-bank private sector.27 In particular,

Levchenko and Mauro (2006) find that FDI is more stable than other types of

flows during sudden stops of capital flows. Also of focus in this literature is the

question of whether internal or external factors are the main drivers of FDI.

On the one hand, Edwards (1991) shows that government size, political sta-

bility, and openness play an important role. On the other hand, Albuquerque,

Loayza, and Serven (2005), for example, analyze the dependence of FDI on

global factors or worldwide sources of risk. In a cross-country time series

data set covering 94 countries and 29 years, the authors show that developing

country exposure to global factors has dramatically increased. This analysis

26While the focus of this overview is on the literature on macro economic determinants ofFDI, another literature convincingly analyzes determinants of FDI from a finance perspec-tive. See, for example, Schnitzer (2000).

27See Marin and Schnitzer (2006) for an insightful analysis on the question, when FDI iscounted as a capital flows at all.


delivers strong support for the hypothesis that increased market integration

leads to increased worldwide sources of risk.

Literature on emerging country vulnerability has shifted focus. In the early

nineties the focus of the empirical literature on capital flows was on the surge of

capital inflows into emerging markets and the implied potential vulnerabilities.

Later, the focus shifted to the question why the capital flows from rich to poor

countries have such a low volume. This more recent literature analyzes which

theoretical explanations of the Lucas paradox28 are relevant. The theoretical

literature has determined differences in fundamentals such as government poli-

cies and institutions’ or international capital markets’ imperfections such as

sovereign risk and asymmetric information.

A prominent example of this literature is Kraay, Loayza, Serven, and Ven-

tura (2005). The authors first contribute to the theoretical literature by mod-

eling North-South capital flows, taking into account the interplay between

diminishing returns at the country level, production risk, and sovereign risk.

The authors then generate country portfolios and a world distribution of cap-

ital that resembles those portfolios actually observed in the data. In contrast

to Kraay et al. (2005), Alfaro, Kalemli-Ozcan, and Volosovych (2005) focus

on the empirical analysis exclusively. With the help of cross-country regres-

sions for the period of 1971-1998, the authors show that institutional quality

is the most important causal variable in explaining the low amount of flows.

Additionally, they find that differences in human capital and asymmetric infor-

mation play a significant role. Reinhart and Rogoff (2004) emphasize capital

market imperfections. They show that serial default among debtor countries is

a usual phenomenon. According to the authors, a key explanation for the low

volume of capital flows from rich to poor countries is the difficulty in borrowing

for countries with a history of defaults.

28Lucas (1990) compares the United States and India in 1988 finding that, if the neoclas-sical model were true, the marginal product of capital in India should be about 58 timesthat of the United States. In the face of such return differentials, all capital should flowfrom the United States to India. However, such flows are not observed. Lucas questionsthe validity of the assumptions that give rise to these differences in the marginal product ofcapital and asks which assumptions should replace these.


Empirical Analyses of Discrete Crisis Events

Although, in principal, sudden stops of capital flows are simply large negative

variations of capital flows that occur in a short time period, the literature

on these pronounced crises events is distinct from the literature on capital

flows. Although observers see a large overlap of variables that explain con-

tinuous changes in capital flows with variables that explain sudden stops of

capital flows, observations also note additional forces at work that trigger the

discontinuous switch in case of a crisis.

A first branch of the empirical literature on sudden stop crises is descrip-

tive. Apart from dating the crisis events, this literature is concerned with

distinguishing sudden stops from other crises such as current account reversals

or currency crises. Calvo and Reinhart (2000) are among the first to give an

overview of sudden stop incidences. The authors present a selection of large

reversals in net private capital flows calculated as percent of GDP in the re-

spective countries. They further illustrate that many countries experiencing a

sudden stop of capital flows in the second half of the 90s witnessed large surges

in capital inflows in the first half of the decade. Furthermore, the authors as-

sess the severity of the crises by analyzing the coincidence of currency crisis,

banking crises, and financial account reversals for a sample of 20 countries

from 1970 to 1994. The authors conclude that the severity, specifically the

huge burden of bailing out the banks as well as the orders of magnitude of the

capital account reversals, has increased during the recent crises periods. While

the paper gives a good introduction into the topic, the criteria of selection of

the included capital flow reversals are not evident. This absence of selection

criteria renders the discussion of policy implications less convincing.

In a series of papers, Edwards (2004b), Edwards (2004a) and Edwards

(2005), the author systematically analyzes the occurrence of current account

reversals and sudden stops of capital flows. Using a panel data set for 157

countries, Edwards (2004a) finds that from 1970 to 2001 there is a 5.6 percent

chance of sudden stops; the chance of current account reversals is 11.8 percent.

The occurrence of the two crises events is highly correlated. Nevertheless,

many sudden stops have not been followed by current account reversals. This

indicates that, when facing a sudden stop, many countries effectively use their

international reserves to avoid an abrupt current account adjustment. At

the same time, a number of countries endure major current account reversals

without facing a sudden stop in inflows. Most of these countries were not


subject to a large surge of inflows to begin with and had financed their large

deficits by drawing down international reserves.29

Due to the close relation between the two crisis phenomena, many explana-

tory variables of current account reversals are also relevant in understanding

the occurrence of sudden stops in capital flows. Analogously to the drivers of

capital flows, the drivers of current account reversals can be categorized into

domestic and external drivers. One of the most prominent empirical inves-

tigations on determinants of current account reversals is Milesi-Ferretti and

Razin (1998). The authors run a panel probit analysis covering 86 low and

middle income countries over a time span from 1974 to 1990. Milesi-Ferretti

and Razin (1998) find the following major determinants of current account re-

versals: domestic variables such as the past current account balance, openness

(share of exports and imports to GDP), and level of reserves, and external

variables such as terms of trade shocks, US real interest rates, and growth in

industrial countries. Using a larger data set, Edwards (2005) additionally finds

the external debt to GDP ratio, domestic credit creation, and debt services

crucial in explaining current account reversals.

In a series of papers Calvo et al. (2004), Calvo et al. (2006), and Calvo,

Izquierdo, and Loo-Kung (2006) analyze the key explanatory factors of sudden

stops of capital flows. The main focus of these papers is on the role of balance

sheet effects, i.e., the interaction of the leverage of the absorption of trad-

ables (i.e., vulnerability against real exchange rate fluctuations) and liability

dollarization, which directly follows from the theoretical considerations in the

theoretical models on sudden stops. All three papers use a sample of 32 devel-

oped and developing countries for the period from 1990 to 2001. Calvo et al.

(2004) and Calvo et al. (2006) use random effects panel probit estimations

controlling for time fixed effects.

Moreover, in addition to the significant effect of the interaction of the vul-

nerability against real exchange rate fluctuations with the liability dollariza-

tion, the authors find the two variables separately significant. Also important

is that negative terms of trade growth increases the likelihood of a sudden

stop significantly. The authors find that, controlling for dollarization, a fixed

exchange rate seems to increase the probability of a sudden stop. However,

this effect is not robust in all specifications. Other tested variables turn out

not to be significant.

29See Edwards (2004b).


Policy Implications and Shortcomings of Empirical Literature

The policy implications of the empirical literature on capital flows and discrete

sudden stop events are that both domestic and external factors play important

roles. Strategies that lead to a stable macroeconomic environment appear to be

as relevant as policies that help to respond to external shocks, as for example

holding a sufficient amount of foreign reserves. One outcome of the empirical

literature is that balance sheet effects should be at the core of government

strategies to prevent crises, i.e., liability dollarization and vulnerability to real

exchange rate fluctuations should be a major concern. Policies regarding the

nominal exchange rate, i.e., questions of a floating versus a fixed exchange rate,

seem to play a much less important role. Capital inflow controls are analyzed

because of the observation that sudden stops are often preceded by periods

of surges of capital inflows. However, Montiel and Reinhart (1999) find that

these controls are only effective in altering the composition of capital flows,

not their magnitude.

Although the empirical literature on sudden stops provides various valu-

able insights, this same literature is also subject to a number of drawbacks.

The first drawback is the lack of analysis of the role of uncertainty about the

fundamentals.

A second drawback of a particular part of this empirical literature is the

following: Calvo et al. (2004), Calvo et al. (2006) and Calvo et al. (2006) use

panel probit estimations with random effects. The estimates will be biased

in the case that there are systematic differences between the countries that

contribute to the explanation of the occurrence of crises. Due to the limited

number of countries within the sample, justifying the use of random effects es-

timation is difficult. The use of pooled probit and logit estimations, controlling

for country and time fixed effects by introducing country and time dummies,

is a more appropriate estimation approach. A third drawback is the use of

yearly data, which leads, partly, to a small number of observations.

1.2.5 Contribution of this Study

This study contributes to both the theoretical and the empirical literature

on sudden stops. First, here follows a discussion of the contribution to the

theoretical literature, succeeded by a description of the contribution to the


empirical literature.

The theoretical part of chapter 2 contributes to the sudden stop models

based on multiple equilibria. The first contribution consists of addressing the

topic of investor coordination. The model reformulates the setup in Calvo

(2003) as a coordination game, thereby addressing the second criticism of the

existing literature. When analyzing the problem from the angle of a coordina-

tion game, it becomes clear that, in an intermediate region of the fundamentals,

a crisis can be triggered by a coordination failure in the case that the true level

of the fundamentals is common knowledge, i.e., the crisis occurs merely be-

cause the investors cannot coordinate on the possible good equilibrium because

each investor fears that the other investors abstain from investing.

In a further step, the model in chapter 2 introduces private information

about the fundamentals. This step helps to confront the first and the third

criticism of the existing sudden stop models based on multiple equilibria. The

introduction of private information, i.e., the application of the global game

approach, leads to a unique threshold equilibrium.30 This, in turn, makes pos-

sible predicting the influence of the variables in the model on the probability

of a crisis. In this context, both internal and external factors can be examined:

the effect of internal factors such as productivity, represented by one model

parameter, and the effect of external variables such as the international inter-

est rate. Then, this work can show that within this model, an improvement

in the internal factors reduces the crisis probability, while an increase in the

international interest rate increases the crisis probability. The most interesting

result stems from the analysis of uncertainty about the fundamentals. If this

uncertainty increases, the probability of a crisis increases as well. This finding

about what can drive crisis probability has policy implications for informa-

tion dissemination strategies of governments. Governments should try to help

private investors in receiving precise private information on the state of the

economy.

Although the theory of global games has been prominently applied to var-

ious different crises phenomena, specific application of this approach is worth-

while for the problem of sudden stops. First, this modeling approach allows

predictions of crisis triggers. Second, it is not clear ex ante that the effect

of the uncertainty about the fundamentals on other crisis phenomena trans-

30The literature on global games and the analysis of uncertainty about the fundamentalsis reviewed in more detail in section 1.4.

1.3. THE SPREAD OF FINANCIAL CRISES 21

lates directly into its effect on sudden stops of capital flows. The strategic

complementarity in the current model arises through a different channel, in

particular a shared tax burden, than, for example, in models of speculative at-

tacks. There, the sum of the funds put to speculation must exceed reserves that

the central bank can use to defend its fixed exchange rate. From a modeling

perspective, the global game approach is also interesting because, as opposed

to most applications where the terms containing random variables enter the

optimization of the agents in the model additively, in this present study model,

the random variable terms enter as multiplicative, which leaves the solution

more challenging.

The empirical part of chapter 2 contributes to the empirical sudden stop

literature. The first contribution of this study is the provision of the missing

empirical analysis of the effect of uncertainty on sudden stop crises. Care has

been taken to ensure that this effect is distinct and present when alternative

explanative factors of sudden stops are controlled for. Thereby, this study

builds on the existing literature and the drivers of sudden stops that have

been identified.

Second, the study improves the existing empirical analyses by Calvo et al.

(2004) by dropping the assumption of random effects. Instead, this study

introduces dummy variables to control for country and time fixed effects in

probit and logit regressions. A rigorous robustness analysis validates the results

found in the benchmark setting.

Finally, this study extends the existing literature by constructing a rich

monthly data set. This allows for two improvements of the existing literature.

Firstly, the small sample size objection is addressed. Secondly, monthly data

allows for a better consideration of the problem of endogeneity resulting of

potential causality running from the crisis to its potential triggers.

1.3 The Spread of Financial Crises

The literature on the spread of financial crises or ”contagion” is as extensive as

the one on financial crises.31 Therefore, this section places particular emphasis

on the literature that is relevant for chapter 3 of this study. In particular is the

31The terms ”spread of financial crises” and ”contagion” are used interchangeably through-out the study.


mention of those articles that build on the insight that common creditors are

at the core of transmitting crises across countries. In addition to providing an

overview of the relevant literature, this section puts this study in perspective

and defines its scope in the context of the literature.

The section is organized as follows: First, the phenomenon of contagion is

defined. Second, the two broad theoretical approaches to modeling contagion

are presented. Third, contagion channels and mechanisms are considered.

Finally, the contribution of this study is considered in light of the relevant

literature.

1.3.1 The Contagion Phenomenon

The term contagion has been defined in a number of different ways in the

literature. While a few researchers define contagion as an increase in the co-

movement of the financial indicators of different countries during crisis periods

that is unexplained by common shocks, other researchers define contagion

simply as co-movement during crisis periods that is unexplained by common

shocks.32 In this analysis, following Didier et al. (2006), contagion is defined

as the propagation of crises across countries beyond what would be implied by

common shocks. The crises analyzed in the chapter on contagion are significant

drops in stock market returns.

The spread of different crisis phenomena has been treated separately in the

literature. While in the beginning the focus was on the spread of currency

crises, research shifted toward questions about the spread of banking crises

and stock market crises. Prominent analyses of the contagion of currency

crises are Eichengreen et al. (1996), Glick and Rose (1999), Drazen (2000),

and Fratzscher (2003). The spread of banking crises has been analyzed by

Dasgupta (2004). The literature on the contagion of stock market crises is

extensive. Prominent examples are Goldstein and Pauzner (2004) or Broner

et al. (2006).

In line with, for example, Broner et al. (2006), chapter 3 of this study

focuses on the spread of stock market crises. Two main considerations explain

this scoping decision. First, the empirical disaggregate analysis of capital

flows, reported in the balance of payments statistics, shows that in particular

32See Rigobon (2003), Forbes and Rigobon (2001), and Kaminsky and Reinhart (2000)for details.


portfolio investments (i.e., cross border stock market investments) and other

investments (which comprise bank flows) are volatile during crisis periods.33

Therefore, questions about how stock market crises and banking crises spread

appear particularly relevant.

Second, especially with regard to an empirical analysis of contagion, the

spread of stock market crises appears more promising than an analysis of the

spread of banking crises. This is due to better data availability of stock market

data and therefore a more reliable identification of crises events.34

1.3.2 Modeling Approaches to Contagion

The theoretical literature on contagion can be distinguished into two main

threads from a methodological standpoint: A first thread of literature is based

on portfolio optimization while the second thread uses models of sequential

investment games.35 In the first thread of literature, the optimal allocation of

investments across countries is done according to considerations of the expected

portfolio return versus the accompanying risks, i.e., the variance of the portfolio

return.36

In the second thread of literature, the game-theoretical investment games

begin with describing a self-fulfilling crisis in the first country. The crisis oc-

curs due to the condition that expected returns of holding an investment until

maturity depend positively on the fraction of other agents doing the same,

whereas an early withdrawal generates a lower but certain payoff. If every-

one expects everyone else to withdraw early, the crisis fulfills itself. Private

information is introduced into this setting to be able to determine a unique

threshold equilibrium dividing the fundamentals into ranges where the attack

versus the non-attack equilibrium exist. In turn, the crisis in the first country

influences the behavior of agents in such way that a self-fulfilling crisis in the

second country becomes more probable. Goldstein and Pauzner (2004) show

that investors become more risk averse in country two, because they lose part

33See Levchenko and Mauro (2006).34See Caprio and Klingebiel (2002) for a comprehensive overview of banking crises. Even

in this comprehensive and elaborate data set on banking crises, the authors refrain fromstating exact crises months.

35The portfolio optimization literature originates from Markowitz (1959). The investmentgames literature is based on Diamond and Dybvig (1983).

36Prominent literature on portfolio optimization are, for example, Kodres and Pritsker(2002), Calvo and Mendoza (2000), and Broner et al. (2006).


of their wealth in the first country.

Both approaches of modeling generate valuable insights into how contagion

functions and, hence, also into preventative policies . These policy implications

will be explained in the next subsection.

The present analysis builds upon sequential coordination games. This

choice stems from my interest in uncertainty about the fundamentals. The

sequence of investment games is an optimal setting to introduce private in-

formation about the fundamentals and to analyze the effect of dispersion of

private information on the vulnerability of a country to contagion.

1.3.3 Contagion Channels

The literature on contagion arrays a variety of explanations on how a crisis in

one country can spread to other countries. Following Hernandez and Valdes

(2001) and Dornbusch, Claessens, and Park (1999), this overview distinguishes

three different explanations: macroeconomic similarities, trade effects, and

financial linkages.

The first part of literature has identified macroeconomic similarities as

main source of contagion. In these models, contagion either occurs because

countries with bad fundamentals are subject to common negative shocks (e.g.,

a rise in US interest rates) or because investors treat all countries that ”look

alike” equally. This behavior of investors can be explained by incomplete infor-

mation. Hence, if a country is hit by a crisis, information spill-overs materialize

against countries in similar situations. The crisis in the first country serves as

a wake-up call according to this view.37

Sachs, Tornell, and Velasco (1996) have analyzed the effect of macroeco-

nomic similarities empirically. They identify the characteristics of countries

performing worse after the Mexican crisis in 1994-95. They find that both

the initial real exchange rate overvaluation and excess bank credit creation

contribute largely to the after-crisis cross country performance. These results

suggest that contagion is driven by initial macroeconomic fundamentals. In

terms of policy implications, this means that the best way of avoiding conta-

gion is for a country to improve on its macroeconomic fundamentals.

The second part of the literature has identified trade linkages as main

37This idea has been put forward by Goldstein (1998).


source of contagion. Direct trade linkages and trade competition have been

analyzed in this context. Trade competition leads to a transmission of crises

through a competitive devaluation: If a trading partner or competitor experi-

ences a strong devaluation, the government may also devalue in order to stay

competitive. Investors foreseeing this step cut demand for the second country’s

assets, thereby triggering a crisis, followed by a devaluation, and therein val-

idating their own expectations. This indirect channel is particularly relevant

in the spread of currency crises.38

Eichengreen et al. (1996) analyze contagion of currency crises in a group

of 20 OECD countries. They define contagion as an increase in the likelihood

of a crisis in a particular country, given a crisis in another country. They

conclude that trade linkages play a more relevant role in explaining contagion

than do macroeconomic similarities. In line with Sachs et al. (1996), Glick and

Rose (1999) try to explain cross-country performance after particular crises.

The authors find that trade linkages are more relevant than macroeconomic

characteristics. If the trade channel is the most relevant in transmitting crises,

the best policy for a country to avoid contagion is to diversify its trade across

trading partners and sectors.

The third part of literature has identified financial linkages as main

drivers of contagion. Apart from direct links through cross-country invest-

ments, the financial linkages occur due to common creditors. Because the

most recent empirical literature supports the view that common creditors are

the most relevant driver of contagion, a number of theories have been devel-

oped to explain the exact mechanism of the spread of crises through common

creditors. These mechanisms are explained in the next section.

The empirical literature supporting the importance of common creditors as

contagion channel started with Kaminsky and Reinhart (2000). The authors

follow a similar strategy as Eichengreen et al. (1996) but use a larger sample

of countries and a different crisis criterion. They analyze the effect of crises in

alternative clusters of countries on the likelihood of a crisis occurring in coun-

tries of that same cluster. They conclude that financial links are potentially

the more important transmission mechanism. However, the confinement is the

high correlation between trade and financial links which makes an ultimate

distinction difficult. Van Rijckeghem and Weder (2001) incorporate a measure

of competition for bank funding between potentially-affected countries and the

38See Gerlach and Smets (1996) for a model of contagion through trade linkages.


initial-crisis country as explanatory variable for the probability of a crisis in

a potentially-affected country. They find that competition for bank lending

is a more robust indicator for incidences of contagion than trade linkages and

countries’ macroeconomic characteristics. Caramazza, Ricci, and R. (2004)

analyze the contagion of currency crises in a panel probit analysis with 41

emerging economies. Controlling for the role of domestic and external funda-

mentals, trade spill-overs, and financial weaknesses, they find a strong effect

of financial linkages to the initial-crisis country. Financial linkages are defined

according to the same measure of competition for bank funding as in Van Ri-

jckeghem and Weder (2001). Hernandez and Valdes (2001) show empirically

the importance of financial linkages for contagion of stock market crises.

The policy implications from the common creditor channel largely depend

on the exact mechanism through which the crisis is spread. Therefore, I men-

tion them in the next section.

1.3.4 Contagion through Common Creditors

Based on the insight into the role of common investors, the theoretical liter-

ature suggests different transmission mechanisms. The most relevant are the

following five: transmission of crises through financial market practices, infor-

mation asymmetries, changes in risk aversion, wealth effects, and information

revelation.

The transmission mechanisms based on financial market practices ad-

dress optimization behavior of financial investors. A prominent example of this

thread of literature is Calvo and Mendoza (2000). The authors show contagion

as a result of the interaction between short-selling constraints, fixed costs of

information acquisition, and fund managers’ performance schemes. Contagion

is defined as a situation in which utility-maximizing investors choose not to pay

for information relevant for their portfolio decision or in which investors op-

timally choose to mimic arbitrary ”market” portfolios. The investors become

susceptible to country-specific rumors because of their lack of information. In

their model of mean-variance portfolio optimization, the portfolio manager’s

performance is measured relative to a market portfolio. Assuming further that

the marginal cost of yielding a below market return exceeds the marginal gain,

in the opposite case, it can become a rational decision to mimic the market

portfolio. If a rumor sets the return of a specific country to be lower than the


market portfolio, then investors might simply follow the herd.

Disyatat and Gelos (2001) empirically validate the theoretical findings by

Calvo and Mendoza (2000). They show that the asset allocation of emerging

market funds can be well approximated by short-sell constraints and mean

variance optimization around benchmark indices.

The policy implication from these studies is the necessity to reassess the

utility from short-selling constraints in the presence of fixed costs of informa-

tion acquisition and the regulations regarding average returns for institutional

investors.

Another article addressing contagion through financial market practices is

Schinasi and Smith (2000). The authors focus on the rebalancing of portfolios

within a mean variance or a value at risk (VaR) framework. In the case that the

investment position of the portfolio is partly financed by debt, i.e., is leveraged,

the loss in one high-risk asset can lead to a withdrawal from other high-risk

assets. To reestablish the optimal portfolio weight after the loss in one high-

risk asset, market participants shift from the low-risk to the high-risk asset.

However, in the case of a leveraged position, the low-risk asset is a negative

position. Hence, less leverage also implies less investment into the high-risk

asset.

The transmission of stock market crises through information asymme-

tries has been prominently analyzed by Kodres and Pritsker (2002). In their

model, differently informed market participants transmit idiosyncratic shocks

across countries by optimally rebalancing their portfolios’ exposure to macroe-

conomic risks. Contagion is defined as a price movement in a market not

initially hit by the shock. The transmission of shocks is induced by rational

investors who are differently informed. Price movements in the second market

are exaggerated because order flows by the informed investors are misinter-

preted as information-based by the uninformed investors. In terms of pol-

icy implications, Kodres and Pritsker (2002) suggest removal of informational

asymmetries to diminish vulnerability to contagion.

The transmission of stock market crises through an increase in investors

risk aversion has been analyzed by Broner et al. (2006). In a mean-variance

optimization setup, the authors illustrate the effect of increased risk aversion

of investors who performed relatively badly due to overexposure to a crisis

country. In this setup, investors hold different portfolios due to heterogenous

beliefs about expected dividends. Investor utility is a function of their own


wealth and their relative wealth to other investors, which depends on their

relative performance. If an investor is overly optimistic with respect to a

specific country and then under-performs there, his risk aversion increases and

he will rebalance his portfolio towards the average portfolio. To do this, he

withdraws from all other countries in which he was formerly overexposed. This

leads to the crisis being transmitted between countries that share overexposed

investors. The model is convincing and in line with empirical observations.

However, the increase in the risk aversion is not explicitly modeled.

In the empirical part of the paper, the authors examine the effect of gains

and losses on investor behavior in terms of portfolio choices. The authors fo-

cus on the Thai (1997), the Russian (1998), and the Brazilian (1999) crises.

They show that under-performing funds do adjust portfolios in direction of

the average portfolio. The presence of overexposed investors, measured by

a time-varying index of financial interdependence, helps explain stock mar-

ket co-movements across emerging markets above and beyond trade linkages.

Also important is a negative correlation between countries’ stock market per-

formance during crises and the degree to which these countries shared overex-

posed funds with the original crisis country.

The analysis by Broner et al. (2006) generates an interesting policy im-

plication. Apart from the sound analysis of the fundamentals of an economy,

the authors suggest close monitoring of the micro-composition of investment

positions across investment funds to detect dangers early.

A number of papers have analyzed the transmission of crises through a

wealth effect. Prominent examples are Goldstein and Pauzner (2004), Kyle

and Xiong (2001), and Yuan (2004). Goldstein and Pauzner (2004) model a

sequence of self-fulfilling roll-over crises (in the sense of Diamond and Dybvig

(1983)) in two countries that have independent fundamentals but are linked

through common investors. The transmission functions as follows: First, a self-

fulfilling roll-over crisis happens in the initial-crisis country. In this crisis, some

investors loose wealth. Investors’ utility function displays decreasing absolute

risk aversion, which implies that investors became more risk-averse after the

loss of wealth. Withdrawing early in any country is the safe action, whereas

holding the investment until maturity is risky. In the latter case, the return

depends on the fundamentals and the action of agents. As a result, agents

in the second country will coordinate on maintaining their investments over a

smaller range of fundamentals. Hence, a crisis there becomes more likely, i.e.,


there is contagion. This generates a positive correlation between the returns

on investments in the two countries, reducing the benefits of diversification.

Goldstein and Pauzner (2004) deliver ambiguous policy implications. De-

pending on the welfare implications of portfolio diversification, on the one

extreme, introducing capital controls can help the situation (when partial di-

versification generates a higher welfare than full diversification). On the other

extreme, subsidizing diversification is the better move (when partial diversi-

fication generates a lower welfare than full diversification). A shortcoming of

this approach is that determining which of the scenarios is closer to the real

world is not possible.

One example of the transmission of currency crises through the revelation

of information in the initial crisis is the model by Taketa (2004). The author

models crises of speculative attacks a la Morris and Shin (1998) in two countries

having independent fundamentals but being linked through common investors.

In the model, two types of investors are characterized by different costs of

attacking the currency. The investor type is private information. If a crisis

reveals other investors’ types, this case will lead each investor to update his

beliefs and thereby change his optimal behavior, which can lead to a crisis in

another unrelated country. Namely, if the fundamentals in the country of the

original crisis country were good, the occurrence reveals the presence of many

investors with low costs of attacking. Hence, the crisis makes investors revise

their beliefs, leading them to attack the second country already in better state

of fundamentals than without the crisis in the first country. This implies that

crises can be the more contagious the better the fundamentals of the crisis

country are.

Although this literature has generated a good understanding of the mech-

anisms through which financial crises spread and many plausible and helpful

policy implications, analysis of uncertainty about the fundamentals is missing.

Additionally, in terms of the empirical analysis, the main channels have been

tested, i.e., whether it is macroeconomic similarities, trade linkages, or finan-

cial linkages that are at the core of contagion. However, only few studies take

on a specific mechanism.



The contribution of this study to contagion literature is the illustration of an

additional mechanism through which financial crises spread in a simple model.

The idea that the dispersion of private signals in potentially-affected countries

increases after a surprise crisis in an initial-crisis country and decreases after

an anticipated crisis has not been modeled so far.

The focus of chapter 3 is on the empirical analysis. The present analysis

contributes to the empirical contagion literature by examining one additional

mechanism through which a stock market crisis can spread.

Additionally, the construction of a rich, monthly data set, spanning from

beginning of 1993 until the end of 2005, allows for improvement on the existing

literature. Firstly, the long time span and cross-sectional dimension allows for

covering the crises in Thailand (1997), Russia (1998), Brazil (1999), Turkey

(2001), and Argentina (2001). Most of the above mentioned papers work with

a collection of only three of these crises. Secondly, this study controls for a

large range of alternative contagion channels adapting the existing measures

of linkages to the specific needs in the context of the spread of stock market

crises.39

1.4 The Effect of uncertainty about the fun-

damentals

The effect of uncertainty about the fundamentals in the context of financial

crises is interesting from two perspectives. From one vantage, observers can

study how far uncertainty is a factor in explaining the occurrence of a crisis

within a country. From another vantage, observers can examine which role un-

certainty about the fundamentals plays in the transmission of financial crises

across countries. A first brief section explains the concept of uncertainty used

in this study while the two subsequent sections mirror these two perspectives.

The first subsection clarifies the concept of uncertainty, while the second sub-

section reviews the literature on the effect of uncertainty on the occurrence

of financial crises. Finally, the third subsection reviews the literature on the

39For example, Broner et al. (2006) among others use the exact same measures of com-mon creditors to control for this channel as other authors use for currency crises, therebyunderestimating the real effect of those alternative channels.

1.4. THE EFFECT OF UNCERTAINTY 31

effect of uncertainty on the spread of crises. This entire section is brief, be-

cause many of the relevant papers have already been reviewed in the sections

on financial crises and on the spread of crises and, therefore, are not repeated

here.

1.4.1 The Concept of Uncertainty

The term uncertainty in the present study refers to the uncertainty about

the fundamentals, which is relevant in the literature of global games. Global

games are coordination games, in which the fundamentals and the information

that agents in the model receive about the fundamentals are random variables

with specific distributions. Following, for example, Morris and Shin (1998),

Heinemann and Illing (2002), or Prati and Sbracia (2002), this study defines

uncertainty about the fundamentals as the dispersion of private signals around

the true value of the fundamentals. In models using global game theory, a cru-

cial assumption is that the true value of the fundamentals is not observable by

agents and that, therefore, its realization is not part of the common knowledge

of the game. Instead, each agent in the game receives a private signal about

this true value of the fundamentals. The dispersion or the noise of the private

signals around the true value of the fundamentals is called uncertainty about

the fundamentals. An increase in the dispersion of private signals is equal to

a decrease in the precision of private signals. In this study, uncertainty about

the fundamentals is interpreted as the dispersion of the agents’ private opin-

ions or agents’ disagreement about the state of the economic fundamentals. In

chapter 2, these economic fundamentals refer to the level of government debt,

while in chapter 3 to the investment environment.

Achieving clarity in this study requires distinguishing the concept of uncer-

tainty about the fundamentals from related concepts. As stated in Heinemann

and Illing (2002), in general, economists identify two kinds of uncertainty: un-

certainty about the fundamentals of the economy and strategic uncertainty.

Strategic uncertainty refers to uncertainty about the behavior of other agents.

While these concepts are closely related, the present study focuses on the un-

certainty about the fundamentals.

A concept closely linked to uncertainty about the fundamentals is trans-

parency. Transparency refers to the precision and timeliness of information


disclosure by public authorities.40 In most analyses transparency is modeled

such that, in all states, increased transparency reduces the dispersion of private

signals around the true value of the fundamentals.41

A number of papers assume that, in addition to private signals about the

fundamentals, public signals exist.42 Public signals are not equal to informa-

tion disseminated by public authorities but are signals observed by all agents

in the model in exactly the same way (they are therefore common knowledge).

Public signals represent the true value of the fundamentals plus a noise term.

In this study, public signals are abstracted from, following the argument that

private interpretation of available information ultimately determines the ac-

tions of investors.

Moreover, readers should distinguish uncertainty about the fundamentals

referred to in this study from volatility over time, which is partly subsumed

under the notion of uncertainty.43

1.4.2 Uncertainty, Crises and their Spread

Carlsson and van Damme (1993) developed the theory of global games by first

introducing private, noisy information into the setup of a coordination game

on investment. This type of model is a natural framework for studying the

role of uncertainty in the occurrence of financial crises because global game

models allow consideration of the effect of changes in the precision of private

and public information on the likelihood of a crisis. Morris and Shin (2003)

provide a concise summary of the global games approach.

This approach has been applied to various setups of crisis models. The

particular appeal here is the following: Applying the global games approach to

models characterized by a multiplicity of equilibria under complete information

serves as an equilibrium selection mechanism; under certain assumptions, the

approach allows removal of the indeterminacy of equilibria completely.

40See Morris and Shin (2004) for this definition.41See, for example, Cukiermann and Meltzer (1986) or Heinemann and Illing (2002).42See, for example, Metz (2002).43See, for example, Mondria (2006b) for a paper closely linked to this study but, among

other differences, uses this different notion of uncertainty. Additionally, much of the financeliterature, for example, uses macroeconomic uncertainty as a synonym for macroeconomicvolatility, a time series based concept. In this context, only recently has the value of surveydata been discovered. See, for example, Giordani and Soederlind (2003) or Arnold and Vrugt(2006).


Additionally, models using the global games approach combine features of

first and second generation models. The equilibrium outcomes result from of

the interaction of levels of the fundamentals and the behavior of agents. Agent-

behavior, in turn, is a function of the expectations of the agents regarding the

fundamentals and the actions of all other agents. In this sense, these models of

crises are both belief- and fundamental-driven at the same time. They belong

to the third generation of crises models. This study benefits especially from

the global games approach, allowing the study of the likelihood of a crisis as

a function of the precision of private signals. Recall that the purpose of this

study is the evaluation of the effect of uncertainty on the occurrence and the

spread of financial crises from a theoretical and an empirical angle.

Theoretical Analyses of Uncertainty

The literature of crisis models applying the global games approach to a setup

where initially multiple equilibria exist is large. This literature starts with

Morris and Shin (1998) who model the occurrence of a currency crisis induced

by a speculative attack. They assume uniformly distributed fundamentals and

uniformly distributed private signals. Finding a unique threshold equilibrium

in terms of the fundamentals, the authors can conduct a comparative static

analysis with respect to the parameters in the model. They show that for

small enough noise of private signals, an increase in transaction costs renders

a currency crisis less probable. Additionally, they show that a higher aggre-

gate wealth of investors, or simply a higher number of investors, increases the

probability of a crisis. However, the authors stop short of analyzing the effect

of uncertainty, despite providing interesting policy implications with regard to

the imposition of transaction costs and capital controls.

Heinemann and Illing (2002) provide a good analysis of the uncertainty

effect in the Morris and Shin (1998) setup. The authors find that a higher pre-

cision of the private signals around the true value of the fundamentals reduces

the likelihood of a speculative attack. In their model, higher transparency

leads to a higher precision of private signals. Hence, the policy implication

of their paper is that governments should provide the best possible private

information but avoid common knowledge in the effort to reduce the risk of

currency crises.


Morris and Shin (2004) apply the global games approach to a model similar

to the Diamond and Dybvig (1983) bank-run model. Assuming normally dis-

tributed fundamentals and private signals, they find that greater precision of

information does not always mitigate the coordination problem.44 Extending

the model used by Morris and Shin (2004), Metz (2002) analyzes the effect of

changes in the noisiness of private and public information on the occurrence

of a currency crisis. She finds that an increase in dispersion of private sig-

nals increases the probability of a currency crisis if the fundamentals of the

economy are sufficiently bad. In this range of fundamentals, an increase of the

dispersion of public information would generate the opposite effect because

then public information serves investors less as a coordination device. The

additional policy implication from this paper is that policy makers must con-

sider that information disseminated to all agents not only affects the outcome

through content but also by functioning as a coordination device.

Papers on the spread of crises, using either the global games approach or

being important for this study because they are linked to uncertainty, have al-

ready been reviewed in section 1.3.1. These are Goldstein and Pauzner (2004),

Taketa (2004), and Mondria (2006a). Global games papers exist on the spread

of banking crises, as for example Dasgupta (2004). However, discussion of

these papers is left to studies dealing with the spread of banking crises.

Empirical Analyses of Uncertainty

Only a few empirical papers explore the effect of uncertainty on the occur-

rence of crises. The existing studies examine the effect of uncertainty on the

occurrence of currency crises. Prati and Sbracia (2002) analyze the effect of

uncertainty on the occurrence of speculative currency crises in the presence of

public and private information about the fundamentals. Bannier (2006) inter-

prets the Mexican crisis as a currency crisis, analyzing the effect of uncertainty

in this context. Both these analyses are based on the theoretical model in Metz

(2002) and assume the presence of public and private information.

Prati and Sbracia (2002) use dispersion of GDP growth forecasts as a mea-

sure combining the effects of public and private signal precision. In their

analysis of seemingly unrelated regressions (SUR) for six Asian countries with

44In this informational setup, the variance of private signals has to be small enough relativeto the variance of the fundamentals in order to be able to find a unique equilibrium.


a monthly data set from 1995 to 2001, they find that dispersion of GDP

growth forecasts has a significant and independent effect from the level of

mean GDP expectations on the exchange rate pressure index. They find a

higher dispersion of GDP growth forecasts increasing exchange rate pressures

when expected GDP growth is above an estimated country-specific threshold,

and forecasts reducing exchange rate pressure when expected GDP growth is

below the threshold.

Some doubts about generality, however, remain about these studies. Ques-

tions on Prati and Sbracia (2002) cluster around the small sample of countries,

application of time series estimations to testing a static global game model,

and finally, estimation of the threshold level to distinguish good from bad fun-

damentals within the main estimation equation. Using the same estimation

method Bannier (2006) is subject to the same criticism.

Tillmann (2004) analyzes the effect of uncertainty on the occurrence of

currency crises in a Markov-switching framework for the French franc and the

Italian lira in 1992. In a framework assuming private information only, he

finds that an increase in the dispersion of private signals renders a speculative

attack more likely. The measure of uncertainty in his analysis consists of

country fund discounts. This is the difference between the price of closed-

end country funds and their underlying net asset value. The results support

the theoretical predictions of Heinemann and Illing (2002), thus enforcing the

policy implications of that paper. One drawback of this analysis is, however,

again, that time series techniques are applied to test a static model.

Based on a model of portfolio optimization, Gelos and Wei (2005) analyze

the effect of different measures of transparency on portfolio holdings. The au-

thors conduct a convincing analysis of the country weights of 137 investment

funds over a time period from January 1996 until December 2000. They show

that both government and corporate transparency have separate and distinct

positive effects on investment flows from international funds into a particular

country. Additionally, they show that during the Thai and the Russian crises

capital flight is greater in the less transparent countries. Hence, as a policy

implication the authors suggest that becoming more transparent is an effec-

tive strategy for countries to benefit from international integration, thereby

avoiding increased volatility.

While including public information in the setup of a global game appears

interesting from a theoretical perspective, keeping track of the empirical rele-


vance renders this concept much less attractive.

Firstly, conveying the meaning of public information is difficult, in partic-

ular if admitting the existence of private information. How does information

emerge that is understood in exactly the same way by all agents of the model

(i.e., information that is common knowledge)? In particular, in the context

of information provided by public authorities, how achievable is a situation in

which all agents understand an announcement in exactly the same way?

Secondly, from a practical view point, available data does not allow for a

reliable distinction between private and public information.

And thirdly, apart from Prati and Sbracia (2002) and Bannier (2006), first

empirical analyses on the effect of uncertainty in the context of financial crises

appear to tend towards a clear-cut effect. This finding seems to contradict

the predictions of a global games setting with public and private information,

where direction of the effect also depend on the level of the fundamentals.

Summing up, these criticisms appear to favor an analysis abstracting from

public information as is done in this study.

To my best knowledge, chapter 3 of this study is the first analysis to test

the effect of uncertainty about the fundamentals on the spread of financial

crises.


The theoretical model in chapter 2 of this study is the first analysis that ap-

plies the global games approach to the question of sudden stops of capital flows.

Much of the literature on global games focuses on currency crisis models a la

Morris and Shin (1998) or on bank-run models of the Diamond and Dybvig

(1983) type. From a theoretical viewpoint, the fact that the strategic comple-

mentarity enters into the model through the shared tax burden is interesting.

Additionally, the multiplicative payoff function is interesting.

The empirical part of chapter 2 contributes to the empirical literature on

the effect of uncertainty on the occurrence of crises by analyzing the effect

of a different type of crisis. The analysis in this study is more rigorous than

the existing literature due to a larger data set. This large data set makes

the exploitation of the cross-sectional dimension of the data possible, which is

more consistent with a static model than the time series analyses mostly done


in the literature.

While the idea of an uncertainty channel of contagion is new, the model

in chapter 3 rather serves illustrative purposes. The focus is on the empirical

part. Again, here, chapter 3 of this study explores a new effect of uncer-

tainty. Especially, this study accomplishes a rigorous empirical analysis of the

existence and the relevance of the effect of uncertainty on the spread of crises.

Chapter 2

Uncertainty About theFundamentals and theOccurrence of Sudden Stops ofCapital Flows: Theory andEmpirics

2.1 Introduction

Most major financial crises involve a sudden stop of capital inflows.1 Examples

are the Latin American debt crises during the 1980s, the crisis experienced in

South East Asia in 1997, and the Russian crisis in 1998.

Recall the example of the Tequila crisis hitting Mexico at the end of 1994

and the beginning of 1995. During this crisis net private capital flows fell by

almost 4 percent relative to GDP in 1994 and a drop of more than 5 percent in

1995.2 The sudden stop was followed by further financial turmoil and severe

consequences in the real economy.

This chapter contributes to the discussion about causes and prevention of

sudden stops in two ways. The first contribution is a model that explains how

1For a list of headline financial crises, see Table 2.12 in Appendix 2.10. These financialcrisis incidents were so severe that they were in the newspaper headlines around the worldand are remembered by most for the associated financial turmoil.

2These percentages correspond to a drop in capital flows of 15.5 billion current US dollarsin 1994 and further 15.2 billion in 1995 in absolute values.

39

40 CHAPTER 2. UNCERTAINTY AND OCCURRENCE OF CRISES

private investors’ uncertainty about the fundamentals increases the probability

of a sudden stop. Uncertainty is interpreted as the disagreement between

the private investors about the quality of the fundamentals.3 The second

contribution of this chapter is to test the predictions of the theoretical model

empirically.

The current theoretical and empirical literature on sudden stops ignores the

effect of uncertainty about the fundamentals, leaving room for a contribution.

It is surprising that the uncertainty about the fundamentals has not been

analyzed in the context of sudden stops of capital flows. In most models,

investors are assumed to take their investment decisions in a forward-looking

manner. They are assumed to base their decisions on expectations about

future returns, which in turn depend on other investor behavior and the future

fundamentals in an economy.4

Whether uncertainty about the fundamentals has an effect on the occur-

rence of sudden stops of capital flows has important policy implications: If,

for example, an increase in uncertainty about the fundamentals increases the

probability of a sudden stop of capital flows, then an economy will be more vul-

nerable in times when uncertainty is higher. Hence, policy makers should take

this fact into consideration. Additionally, to the extent that public authorities

can influence the degree of investors’ uncertainty about the fundamentals, pol-

icymakers should adjust their policies on information dissemination to reduce

uncertainty.

The basic model is an extension of a model by Calvo (2003). As a first

extension, I introduce infinitely many investors of mass one. Then, I set up

a coordination game. In this basic model, investors maximize the value of

their firm, which is the net present value of their after-tax returns net of

investments. The government mechanically sets the tax rate that is necessary

to cover the exogenously given amount of debt. However, the tax lowers the

after-tax productivity of capital, which is the crucial variable in the investment

decision of the investors.

For sufficiently low levels of debt, the government sets output taxes so low

3This interpretation is in line with the global games literature, reviewed in section 1.4.4In the present model, sudden stops of capital flows are assumed to be unexpected to

private investors. In this sense, investors are still not forward looking. Nevertheless, it canbe shown that the results of the analysis also hold for an expected sudden stop, see Calvo(2003).

2.1. INTRODUCTION 41

that investing is attractive. In this model, the level of investment directly

determines the economic growth of the economy. Hence, a high level of invest-

ment induces high growth. On the other extreme, if the debt is high, only low

growth can be observed due to the negative impact of the high taxes.

However, for intermediate levels of the government debt, the optimal action

of a firm depends on the actions of the other firms. If a firm expects all other

firms to invest, it is optimal for this firm to invest as well. This is due to

the debt burden being shared by a large number of other firms. Hence the

government can choose a low tax rate and the after-tax returns of the investing

firms are high. Otherwise, if a firm expects that few other firms will invest,

it is optimal for this firm to abstain from investing. Otherwise the firm will

have to pay high taxes because the debt burden is shared by few investors.

The strategic complementarity between investments that results from the tax

burden being shared among investors explains a multiplicity of equilibria in

the intermediate debt region. High (low) growth induces low (high) output

taxes, which in turn generates high (low) economic growth.

A sudden stop occurs if growth discontinuously switches from high to low

growth. With the help of the methodology of global games, it is possible

to show that there is a threshold level of the government debt, below which

everyone invests and above which no one does. Specifically, the assumption of

private information on the level of debt allows for finding the unique threshold

level. This assumption means that the true value of the government debt is

no longer common knowledge but that every investor receives a private signal

on the level of debt.5

This results in a new equilibrium condition, which permits the determi-

nation of a unique threshold in terms of the debt level: At the threshold

signal, each investor is indifferent between investing and not investing. This is

because at this level of debt the expected payoff of investing, given that each

other investor invests weighted with the conditional probability that each other

investor receives a better signal and hence invests, equals the expected loss of

an investment in the case that each other investor does not invest weighted

with the conditional probability of this event. The probabilities are conditional

on the private signal that the investor receives about the value of the debt.

5It is a plausible assumption that investors interpret published information about thestate of the fundamentals differently. In the current model, it is assumed that governmentdebt and private signals are uniformly distributed.


Due to the monotonicity of the payoff function in the private signal, clearly,

each investor invests if he receives a smaller signal than the threshold signal

and does not invest above. Hence, above the threshold, the economy drops to

the low growth equilibrium due to a lack of investment, although the state of

the fundamentals would still support the high growth equilibrium. The reason

for the switch is a coordination failure between private investors.

Analyzing the threshold equilibrium yields a set of interesting comparative

static results. In the present setup, the change in the value of the threshold

translates into a change in the probability of a sudden stop crisis. The first

result is that the probability of a sudden stop increases with the dispersion of

the private signals on fiscal burden. The second result is that the crisis proba-

bility decreases with the parameter of technological progress. This parameter

can also be understood as an indication of how safe an investment is. The

third result is that the probability of sudden stops increases with the inter-

national interest rate. Finally, the fourth result is that technological progress

and international interest rate influence the scope of government policies.

The policy implications of these findings on sudden stop conditions are

that governments should take uncertainty about the fundamentals into account

because the uncertainty condition has real effects. Two kinds of advice can

be given to governments based on this study. First, governments should help

private investors obtain precise private information on the fundamentals. A

government could achieve this by allowing unrestricted access to government

data for independent institutions. These institutions could then sell the data to

other market participants. One could argue that it would not make a difference

if the government itself sold this information. However, a government could

have incentives to understate the true value of the debt and then ask for higher

taxes ex post. This credibility problem could be alleviated with the help of an

independent institution. As a second policy advice, the results of the study

suggest that governments should care for investment safety in their country

and foster technological progress.

How relevant are the described effects in reality? Do the model and policy

prescriptions apply to real sudden stop cases? Answering these questions re-

quires empirical verification of technological progress, the international interest

rate, and most importantly, uncertainty about the fundamentals influencing

the probability of a sudden stop.

Three hypotheses can be derived from the theoretical model. They build


the basis for the empirical analysis. The first hypothesis is that sudden stops

are less likely if internal factors of an emerging market country become more fa-

vorable, e.g., if the government adopts technology-enhancing policies or takes

measures to ensure investment safety. The second hypothesis is that more

sudden stops occur if the international interest rate increases. The third hy-

pothesis is that a sudden stop is more likely to occur with more uncertainty

on a government’s fiscal policy (less precise private information).

Take the third hypothesis. In the empirical analysis, I focus on the effect

of uncertainty on the probability of a sudden stop. The dependent variable

in the regressions is a dummy variable that takes a value of one in a sudden

stop period and of zero otherwise. In line with Calvo et al. (2004), Cavallo

and Frankel (2004), and Eichengreen, Gupta, and Mody (2006), I identify

sudden stops of capital flows by considering both the first and the second

moments of a measure of capital flows. Provided that in a particular period

the capital flows drop as low as two standard deviations below the sample

mean, a sudden stop crisis period starts when the flows drop lower than one

standard deviation below the sample mean. For symmetry, the crisis period

stops when the flows exceed this limit again. The most important explanatory

variable in the analysis is a measure of uncertainty about the fundamentals: the

variance of investor’ expectations about the fundamentals. The base data are

expectations about GDP growth of the current and following year. These data

are collected by Consensus Economics and the IFO Institute for Economic

Research. I select these growth forecasts because they are available for a

sufficiently large sample of countries.

As a benchmark regression, I use a pooled probit estimation controlling

for country and time fixed effects. The data set contains 31 developing and

developed countries. The sample size is dictated by data availability. The

analyzed period extends from 1990 to 2001, including yearly and monthly

data.

The search for determinants of a sudden stop quickly leads to a problem

of potentially omitted variables and endogeneity. To tackle these difficulties,

I run various robustness checks. Specifically, to address the first problem of

omitted variables, I include a large variety of control variables. To address

the endogeneity problem, I estimate the model with an increasing order of

lags of the explanatory variables. Additionally, I employ two-step estimation

where I instrument the uncertainty in the current period with its own lag.


Furthermore, I check the robustness of my results by conducting the analysis

for the full sample and an emerging market sample by making use of a yearly

and a monthly data set and by using various different estimation methods.

The positive effect of the uncertainty on the occurrence of a sudden stop is

robust across these tests.

Calculating the marginal effects of a one unit increase in uncertainty sug-

gests that an increase of the uncertainty by one standard deviation increases

the crisis probability by up to nine percent.6 These results, indeed, suggest

that the uncertainty about the fundamentals has a non-negligible effect on the

probability of a sudden stop in reality and should thus be incorporated in the

considerations on economic policies.

This study is connected to the following threads of economic literature.

Calvo et al. (2004) analyze drivers of sudden stops, finding that the vulner-

ability to real exchange rate fluctuations and domestic liability dollarization

increase the probability of a crisis. Edwards (2005) focuses on capital mobility

and disputes its link to increased crisis probability. Furthermore, the question

whether internal or external (global) factors drive capital flows into and out

of emerging markets has been extensively studied. Calvo et al. (1993), Calvo,

Leiderman, and Reinhart (1996), Fernandez-Arias (1996), and Montiel and

Reinhart (1999) examine internal factors such as, for example, the price of

debt on the secondary market, country credit ratings, and the domestic rate

of inflation versus external ones such as the interest rates and the economic

activity in highly developed countries. These analyses attribute a higher im-

portance to external factors. In the more recent literature with a focus on

FDI (foreign direct investment), Albuquerque et al. (2005) find that the most

important driver of capital flows is a synthetic global factor, which they inter-

pret as a globalization measure. Broner and Rigobon (2005) detect regional

patterns in capital flows and emphasize the role of contagion in determining

capital movements to a country.

However, the literature ignores the effect of uncertainty about the funda-

mentals on the occurrence of sudden stops. The issue has been addressed so

far only in the context of currency crises. Prati and Sbracia (2002) analyze

the effect of uncertainty on the occurrence of a currency crisis. With their

seemingly unrelated time series regressions (SUR) for six Asian economies,

they show that higher dispersion of GDP growth forecasts (their proxy for the

6See Table 2.8 in section 2.7.2.

2.2. THEORETICAL BACKGROUND 45

fundamentals) tends to have an additional independent effect apart from the

effect exercised by the lagged level of the fundamentals.

The analysis by Prati and Sbracia (2002) suffers from two shortcomings.

The first shortcoming is its application to a small sample of countries all af-

fected by the East Asian financial crisis in 1997-98. Second, given that Prati

and Sbracia (2002) model a static global game, an analysis emphasizing the

cross sectional variation between countries seems more appropriate than the

time series analysis conducted. To overcome these shortcomings in the present

analysis, I work with a much larger set of countries, emphasizing the cross-

sectional variation between countries by estimating pooled probit regressions

controlling for time effects.

2.2 Theoretical Background

This section presents a coordination game model on the occurrence of sudden

stops. The model is based on the framework presented in Calvo (1998b) and

Calvo (2003). I extend the original Calvo setup by introducing a continuum

of infinitely many identical firms of mass one. In a first step, I set up a

coordination game with common knowledge. In a second step, in section 2.4 I

extend the setup further by introducing private signals on the fundamentals.

2.2.1 The Firms

Following Calvo (1998b) and Calvo (2003), each of the infinitely many firms

produces tradable output with a linear homogeneous production function, in

which tradable capital is the only production factor. Capital is fully interna-

tionally mobile ex ante but immobile after investment.

The firms maximize their value by choosing between constant growth paths.

The value of a firm is defined as the sum of discounted future cash flows until

infinity. Due to the linear production function, the rate of investment or capital

accumulation equals the rate of output growth. In their optimization, firms

consider the technology parameter, the tax rate, and the international interest

rate as given. Thereby, the following representation of the value of a firm i


can be found.7

V i =α(1− τ)− zi

(r − zi)(2.1)

V i represents the firm value, α is the productivity factor, τ is the constant

output tax rate, zi is the investment rate that the firm can choose. r represents

the constant international interest rate. Optimizing the value of the firm with

respect to the rate of investment or capital accumulation leads to:

∂V i

∂zi=

α(1− τ)− r

(r − zi)2

The model delivers corner solutions. If the after-tax return on capital,

α(1 − τ) exceeds the international interest rate r, it is optimal for a firm to

invest as much as possible and thus grow as fast as possible. If the return on

capital is lower than the interest rate, the firm does not accumulate capital at

all. Such a firm would even borrow as much capital as possible and invest it

abroad. For the model to deliver a sensible outcome, the parameter zi must

be restrained to finite ’corners’. Following Calvo (2003), the value of zi is

restricted to an interval [0, z] with z < r, in which the lower bound ensures that

capital cannot be unbolted. The upper bound stands for reasonable outcomes

with respect to the valuation of the firms. In particular, as zi signifies the

constant growth path of the firm, by bounding it, I rule out the possibility

that the firm can outgrow the world market in the infinite horizon.

A firm would never invest if this investment had a negative effect on the

value of the firm. Hence, it suffices to consider the sign of the derivative of V i

with respect to zi. If the sign is positive, the agent invests as much as possible

(restricted to z < r in this model); if negative, investment equals zero.

sgn∂V i

∂zi= sgn[α(1− τ)− r] (2.2)

7For a detailed derivation, please see Appendix 2.10. The firms expect the tax rateto be constant, because a sudden stop is unexpected to them. In the light of possiblegrowth collapses and ensuing sudden stops, a different tax policy τt might be optimal forthe government. Therefore, firms would expect the tax rate to change once a crisis occurs.Calvo (2003) shows that growth collapse and sudden stops also occur in the case when theyare foreseen by the firm. Therefore, I do not consider the case of an anticipated crisis here.

2.2. THEORETICAL BACKGROUND 47

2.2.2 The Government

The government inherits a stock of debt D, which must be financed by an

output tax. The tax rate is set so that the future discounted tax revenues

cover the amount of debt. This is possible, assuming full capital market access

by the government.

D = ατ

∫ ∞

0

Kecont e−rtdt =

ατ

r − zecon(2.3)

with

zecon =

∫ 1

0K idi∫ 1

0K idi

=Kecon

Kecon

The superscript econ indicates that a variable refers to the economy and not

to an individual i.

2.2.3 The Reduced Form Game Between Firms

The mechanical way in which the government sets the tax rate introduces

strategic complementarity between the firms into the model. The profit of

investment for an individual company positively depends on the rate of invest-

ment of all other firms. This can be shown by solving Equation (2.3) for τ and

plugging it into Equation (2.2):

sgn∂V i

∂zi= sgn[α−D(r − zecon)− r] (2.4)

The return on investment is a positive function of zecon. This results from

the burden of debt repayment being carried by more firms. Through the tax-

setting mechanism, the investment decision of each firm depends negatively on

the state of the fundamentals.

The main mechanism underlying the interaction of firms is therefore: If

growth is high, the government sets a low tax rate, which in turn sustains

high growth. Similarly, if growth is low, the government sets a high tax rate,

holding firms off investing, which in turn further induces low growth.


2.3 The Common Knowledge Game

First, it is assumed that all the firms and the government know the true values

of the relevant variables. Recalling that firms either do not invest or invest z,

a strategy πi is defined as πi : [D, D] → [0, 1], in which D is the lower and D

is the upper bound of the support of the debt level. This means that firm i

invests in state D with probability πi(D). Because of the mass of agents being

1, the fraction of agents that invest at a particular state of fundamentals can

be expressed as π−i(D) =∫ 1

0πj(D)dj for j 6= i. This makes it possible to

express the investment rate of the economy as the product of upper limit of

investment and the fraction of agents that invest in the economy.

zecon(D) = zπ−i(D) (2.5)

2.3.1 High Growth and Low Growth Equilibrium

Equation (2.4) can be used to illustrate the parameter range for which the

low growth and the high growth equilibrium exist. On the one hand, the low

growth equilibrium can exist if a firm does not have an incentive to deviate

from its strategy not to invest, given that all the other firms do not invest.

This is the case if Equation (2.4) displays a negative value in the case that

zecon = z ∗ 0 = 0. Solving for the debt level yields that the low growth

equilibrium exists in the case that the debt is higher than a threshold:

D > D =α− r

r(2.6)

On the other hand, the high growth equilibrium exists if a firm does not

have an incentive to deviate from the strategy to invest, given that the other

firms do also invest. In terms of Equation (2.4), this means that the high

growth equilibrium exists if the signum of the equation is positive for zecon =

z ∗ 1 = z. Thereby one finds that the high growth equilibrium exists below a

threshold:

D < D =α− r

r − z(2.7)

2.3.2 The Tripartite Classification of Fundamentals

The level of debt can be classified into three ranges. By definition 0 < z < r

and α > r. Therefore, clearly, from the two equations in section 2.3.1 follows

2.3. THE COMMON KNOWLEDGE GAME 49

that D is greater than D. Between the two threshold values, D and D, the

two equilibria coexist. Above D, only the low growth equilibrium exists, and

below D, only the high growth equilibrium exists.

If D is smaller than D, there exists a dominance region of investment. Con-

sequently, the economy will be in high growth equilibrium. If D lies between D

and D, it is not clear if firms can coordinate on the high growth equilibrium or

if a coordination failure occurs and the economy is captured in the low growth

equilibrium. If D exceeds D, there is a dominance region of no investment and

the economy displays low growth with certainty.8 The tripartite classification

of fundamentals is illustrated in Figure 2.1.

D

Dominance Regionof bad fundamentals

Dominance Region of good fundamentalsDominance Region of good fundamentals

DD

Area of possible coordination failure

D*

z

Only high-growthequilibrium exists

Only low-growth equilibrium exists

Multiple equilibria

Figure 2.1: Model Setup

Figure 2.1 shows the existence of the high growth and the low growth

equilibrium as a function of the government debt level. In case of common

knowledge of the true value of government debt, the model displays indetermi-

nacy between the high growth and the low growth equilibrium for those debt

levels, where both equilibria coexist.

8The threshold cases, in which D = D and D = D, are not of interest and will thereforenot be discussed.


2.4 The Private Information Game

Introducing private, slightly noisy information on the state of the fundamentals

makes eliminating the range of multiplicity between D and D possible. Instead,

there is a threshold value of the debt level, below which all agents coordinate

on the high growth equilibrium and above which no agent invests. The dotted

line in Figure 2.1 represents the equilibrium that is realized in the private

information game.

This section explains, first, the informational structure of the model with

private information, second, the optimization of the firms, and third, the proof

of uniqueness of the threshold equilibrium. The next section, then, shows how

the threshold equilibrium is influenced by changes in the technology parameter,

by changes in the international interest rate, and by changes in the precision

of the private signal.

2.4.1 Informational Structure

The agents cannot observe the true debt value but receive noisy signals Di on

the state of the debt. The true debt level is uniformly distributed over the

interval [D, D]. The signals are privately observable and uniformly distributed

in an ε surrounding of the true debt value Di ∼ U [D − ε,D + ε]. The agents

know the distribution and the support of D and of the private signals. All

agents know that all other agents also receive private signals.

The fact that the signal on the state of the debt is private reflects that

agents interpret officially announced values of the government debt differently.

In addition, debt levels are often revised ex post by public authorities. This

enforces the importance of the interpretation of information and justifies the

signals on debt being private.

Deriving a unique equilibrium requires care that the signal is informative

about the true level of debt. Otherwise, the agents would not have any idea

about the true value of debt and about the possible signals that the other

agents receive, given their own signal. As shown in Heinemann and Illing

(2002), the distributional assumptions in the current setup ensure that this

requirement is fulfilled.

2.4. THE PRIVATE INFORMATION GAME 51

2.4.2 Object of Optimization

The firms cannot observe the true value of D in the private information setup

but only have an expectation about it, given the private signal that they

receive.

Because a firm’s expectations of D and of zecon depend on the private signal,

the firm has an expectation about the tax rate that the government will set,

given its private signal:

E(τ |Di) = E

(D(r − zecon)

α|Di

)(2.8)

Therefore, the expectation of the value of the firm depending on the level

of investment can be expressed as:

E(V |Di) = E

(α−D(r − zecon)− zi

r − zi|Di

)(2.9)

The optimizing behavior in the private information game is analogous to the

behavior under common knowledge. Agents maximize the expected difference

in payoffs resulting from alternative strategies: investing versus non-investing.9

However, the expectations of an agent are now conditional on the signal that

he receives. Each agent weighs the expected payoff of investing: for the case

that each other agent invests and for the case that no one else invests, with the

conditional probability given his private signal that the other agents choose

the respective strategies, i.e., investing versus not investing.

Assuming private information, a firm’s strategy is a function of the pri-

vate signal instead of being a function of the true value of the fundamentals:

πi(Di) : [D, D] → [0, 1]. As shown before, the extreme strategies of maxi-

mum investment versus not investing at all dominate all intermediate strate-

gies. Therefore, when calculating the payoff difference, comparing the expected

payoffs of these two strategies suffices.

The investment rate of the economy can be expressed as the product of

the maximum investment z times the fraction of other firms investing, which

depends on their respective signals Dj:

zecon = zπ−i(Dj) = z

∫ 1

0

πj(Dj)dj (2.10)

9In the following, this will simply be referred to as payoff difference. As in Doenges andHeinemann (2001), in the present model also, the payoff of the alternative action dependson the state of the fundamentals and is not fixed to a constant value.


The expected payoff difference P , given the private signal Di, is then

P (Di) = E(α−D(r − zπ−i(Dj))− z

r − z− α−D(r − zπ−i(Dj))− 0

r − 0|Di

)

In the case of unbiased signals around the true value, the expectation of the

true value of a variable, given the private signal that individual firm i receives,

is the signal itself: E(D|Di) = Di. Therefore, the above expression can be

simplified to:

P (Di) = zα− r − rDi + zE(Dπ−i(Dj)|Di)

(r − z)r(2.11)

2.4.3 Unique Equilibrium

This section shows the existence and the uniqueness of the equilibrium. The

analysis proceeds in several steps. First, it is assumed that all agents follow

a simple switching strategy. Second, the starting points for the iterative elim-

ination of dominated strategies are determined. In this process, dominated

strategies are excluded from the set of possible strategies until the unique

threshold equilibrium is found. Third, the monotonicity of the expected pay-

off difference in the private signal is proved. Steps one through three enable

the fourth step, the iterative elimination of dominated strategies beginning at

borders of the dominance regions. Finally, in the fifth step, it is shown that

there is only one unique value of the level of debt for which the payoff differ-

ence given the private signal equals zero. This level of debt is the threshold

value below which all agents invest and above which, no one does.

In the first step it is assumed that each firm follows a simple switching IT .

This means that a firm invests with probability one, if and only if, the signal

it receives is below a threshold T and abstains from investing with probability

one, if the signal is above the threshold: 10

IT =

1 if Di < T0 if Di ≥ T

11 (2.12)

10By continuity arguments, it is possible to show that such a simple switching strategy isoptimal. So generality is not lost when imposing it in the first place.

11In terms of the payoff, the behavior of the firms in a single event is irrelevant. Therefore,it is also irrelevant whether firms invest at Di = T or not.

2.4. THE PRIVATE INFORMATION GAME 53

At this point, assuming a simple switching strategy, the fraction of other

agents investing can be expressed as the probability that one agent receives a

smaller signal than the threshold signal.

Lemma 2.1 Under the assumption that, in the game with infinitely many

agents of mass one, all follow the same switching strategy IT , the fraction of

agents investing, π−i(Dj), can be replaced by the probability that one agent

receives a signal smaller than the threshold signal T , prob(Dj < T |D), in

Equation (2.11).

Proof. See Appendix 2.10.

Given Lemma 2.1, the payoff difference can be expressed in the following

way:12

P (Di, IT ) = zα− r − rDi + zE(Dprob(Dj < T |D)|Di)

(r − z)r(2.13)

The second step is determining the starting points of the elimination of

dominated strategies.

Lemma 2.2 The starting points for the iterative elimination of dominated

strategies are D and D.


The third step shows proof of the monotonicity of the payoff difference

needed as one further ingredient to apply the iterative elimination of dominated

strategies.

Lemma 2.3 P (Di, IT ) is strictly monotonically decreasing in the private sig-

nal Di.


Given Lemma 2.2 and Lemma 2.3, the lowest possible threshold for a

switching strategy of all the firms is D. Similarly, the highest possible thresh-

old is D. For all Di < D, the payoff difference is positive, irrespective of the

12This probability (and the fraction of firms investing) depends on the realization of D.Hence, given private information, firm i’s expectation of the probability is a function of therealization of the private signal Di.


actions of all other firms. As the rationality of the agents is common knowl-

edge, not to invest is a dominated strategy for signals below D. And, at the

other extreme, for all signals Di > D the payoff difference is negative.

Due to the strategic complementarity between investments, the worst sce-

nario that a firm has to consider is the case where IT = ID. This means that

for all values of debt in the range of multiplicity, firms choose not to invest,

although the levels of debt would in case of coordination on the high growth

equilibrium also allow for this. The best scenario would be a switching strategy

of IT = ID.

Steps one through three provide all ingredients to begin the fourth step,

which iteratively eliminates all dominated strategies. This process works as

follows: If a firm i receives a signal that is very close to the border of the

dominance region, the probability that other firms receive signals within the

dominance region and thus have a dominant strategy is very high. Due to the

strict monotonicity, this suffices to induce firm i to have a dominant strategy

as well. This is true for all firms. Therefore, the range between the signal

of firm i and the former border of the dominance region can be added to

the dominance region. The iterative elimination starts at both borders of the

multiplicity range. Starting at the low [high] end of the multiplicity range,

the iterative process continues until finding the maximum [minimum] signal,

at which firm i is indifferent between investing and not investing and which is

at the same time the threshold of the switching strategy of all other firms.13

Milgrom and Roberts (1990) have shown that in all games with strategic

complementarity, the set of strategies that resist the iterative elimination of

dominated strategies is limited by Nash equilibria. Nash equilibria are not

eliminated through this process. Thus ID? and ID? are the most extreme Nash

equilibria of the game. There is no Nash equilibrium below D? in which the

firms do not invest. On the other hand, there is also no Nash equilibrium

above D?

in which the firms invest.

Given this argument and due to the strict monotonicity of the payoff from

Lemma 2.3, it suffices to show that equation

P (Di = D?) = zα− r − rD? + zE(Dprob(Dj < D?)|Di = D?)

(r − z)r= 0 (2.14)

13For a more formal consideration of the iterative elimination, please refer to Appendix2.10.

2.5. COMPARATIVE STATICS 55

has a unique solution. This can be expressed in the following Lemma.

Lemma 2.4 There exists only one value, for which the expected payoff dif-

ference equals 0 given that firm i receives exactly the threshold signal D? as

private signal and given that all other firms have a switching strategy where

the switching signal equals exactly D?.


The unique solution is

D? =α− r − ε

6z

(r − 12z)

(2.15)

D? determined by Equation (2.15) is the unique threshold equilibrium of

the game with private information. This can be summarized in the following

proposition:

Proposition 2.1 There exists a unique threshold equilibrium D? of the game

with private information, such that each firm invests if and only if Di ≤ D?

and does not invest if Di > D?.

Applying the methodology of global games allows for elimination of the

multiplicity range. This, in turn, allows the prediction of the levels of fun-

damentals, at which a growth collapse occurs. In the Calvo setup, a growth

collapse automatically entails a sudden stop of capital flows. So the above

analysis not only lays bare how the economy will plunge into a growth collapse

but, at the same time, explains the onset of a sudden stop of capital flows. It is

of interest to know how the change of economic variables alters the threshold

and, thereby, the probability of a sudden stop.

2.5 Comparative Statics

This section analyzes how a change in the productivity of the country, a change

in the international interest rate, and a change in the noise in the information

on the debt influence the value of the threshold equilibrium at which the growth

collapse and, therefore, the sudden stop take place.


2.5.1 Changes in the Technology Parameter α

First, consider the technology parameter, which in the model is equivalent to

the productivity of capital.

Proposition 2.2 If the technology parameter α increases, the threshold equi-

librium is shifted to a higher debt level, i.e., a growth collapse and, therefore,

a sudden stop occurs at higher debt levels only.

Proof. Differentiating Equation (2.15) with respect to α delivers:

∂D?

∂α=

1

(r − 12z)

> 0 (2.16)

The above expression must always be positive because z < r. This implies

that an increase in α shifts the threshold level to a higher debt level.

Considering the finite support of the distribution of the debt, Proposition

2.2 implies that the probability of a growth collapse decreases and with it the

probability of a sudden stop. In Figure 2.2 this is mirrored by D′? lying right

of D? with α′ being bigger than α.

Another interesting result emerges when looking at the change of the bor-

ders of the multiplicity range with a change in the technology parameter.

Proposition 2.3 If the technology parameter α increases, the range of debt

levels, for which the multiplicity of equilibria prevails, widens in the common

knowledge game.

Proof. The derivative of the lower bound of the multiplicity range, D, with

respect to α is smaller than the derivative of D.

0 <∂D

∂α=

1

r<

∂D?

∂α=

1

(r − 12z)

<∂D

∂α=

1

r − z(2.17)

As illustrated in Figure 2.2, the range of multiplicity enlarges with bigger

α. Between D? and D lies the range, where the low growth equilibrium prevails

due to coordination failure, although in terms of the fundamentals, still, the

high growth equilibrium is possible. One interpretation of this range is to see

it as a measure of inefficiency of the economy. According to this interpretation,


an increasing α implies an increase in the inefficiency of the economy. How-

ever, this view is incorrect. Simultaneous to the increase of the range between

D? and D, also, the range between D and D? increases by the same amount.

For these debt levels, investors coordinate to the high growth equilibrium al-

though, also, the low growth equilibrium exists. Therefore, a more convincing

interpretation of the range between D? and D is the following: It is a range

where the government can improve the situation by helping investors to co-

ordinate. This suggests that technological progress has two positive effects.

Firstly, it directly decreases the probability of a sudden stop and secondly it

accords a larger scope to government policy to enhance coordination.

Note further, that the effect of α decreases in r.14 This can be explained by

the fact that the scope of action for the government is reduced, when external

factors change in an unfavorable way. One example of such a change is an

increase in the international interest rate.

D DD*

~U

D, Di

~U (α)

D‘ D‘D*‘

~U

D, Di

~U (α‘)

with α‘>αD DD*

~U

D, Di

~U (α)

D DD*D DDD*

~U~U

D, Di

~U (α)~U (α)

D‘ D‘D*‘

~U

D, Di

~U (α‘)

D‘ D‘D*‘D‘ D‘D‘D*‘

~U~U

D, Di

~U (α‘)~U (α‘)

with α‘>α

Figure 2.2: Changes in D? and the borders of the multiplicity area due tochanges in α

2.5.2 Changes in the International Interest Rate r

First, I show the direct effect of the international interest rate on the threshold

equilibrium. Next, I show the effect of the international interest rate on the

borders of the multiplicity range in the common knowledge game.

14To see this, differentiate the right hand side of Equation (2.16) with respect to r.


Proposition 2.4 If the international interest rate r increases, the threshold

equilibrium is shifted to a lower level of debt, i.e., a growth collapse and thereby

a sudden stop occurs already at lower levels of debt.

Proof. A change in the international interest rate produces the following

effect:∂D?

∂r=

(3 + ε)z − 6α

6(r − 12z)2

< 0 (2.18)

The denominator of the fraction in Equation (2.18) is always positive. The

numerator could, in principle, be positive or negative. However, given the

assumptions on α and ε, it is always negative. Recall that α exceeds r, which

in turn exceeds z. ε is restricted to be a small positive number. In the limiting

case that α = z, ε has to be smaller than 3 for ∂D?

∂rto be negative. Given the

assumption on the possible size of ε, this restriction is not binding. Hence,

the effect of a change of r on D? is negative. This means, if the international

interest rate increases, D? moves to lower levels of debt, i.e., a growth collapse

already happens at better states of the fundamentals.

In Figure 2.2, this translates into a shift of the threshold to the left if the

international interest rate increases. In terms of the real economy this implies

that, with higher international interest rates, a sudden stop becomes more

probable.

Proposition 2.5 If the international interest rate r increases, the range of

multiplicity of equilibria shrinks in the common knowledge game.

Proof. The derivative of the lower bound of the multiplicity range, D, with

respect to r is bigger than the derivative of D.

0 >∂D

∂r= − α

r2>

∂D

∂r= − α− z

(r − z)2(2.19)

This inequality can be reformulated to:

2αr − αz − r2 > 0

which must hold because α > r > z > 0 by assumption, and hence, αr > r2

and αr > αz.

The comparative static analysis lays bare that the derivative of D with

respect to r is more negative than the one of D. This result implies that the


range of multiplicity shrinks with increasing r. This implies that the scope of

government policies is diminished by an increase in the international interest

rate.

Again, this comparative static result reveals the opposing effects of α, a

parameter determined in the respective country, and r, a parameter, which is

independent of the situation in the particular country.

These opposing effects of α and r are fully in line with the empirical lit-

erature on pull and push factors with respect to capital flows.15 As a large

part of the relevant literature tries to explain the surge of capital inflows into

developing countries, pull refers to the factors that lie inside the economy

and attract capital inflows. Montiel and Reinhart (1999) define these capital

attracting factors as the ones operating through an improvement in the risk-

return characteristics of assets issued by the developing country debtors. Such

an improvement could result from productivity-enhancing economic reforms.16

So in the present setup, these would translate into policies that increase the

technology parameter α.

The most prominent of the push factors (which lie in the industrialized

countries) is the world interest rate.17 In their paper on inflows of capital to

developing countries in the 1990s, Calvo et al. (1996) mention that low interest

rates in developed countries attracted investors to the high investment yields

and improving economic prospects of economies in Asia and Latin America in

the beginning of the 1990s. For example, the short term interest rate in the

United States reached its lowest point since the early 1960s in 1992. Fernandez-

Arias (1996) contributes an interesting twist to the question of the influence of

external factors on capital flows to emerging markets. He shows the positive

effect of lower world interest rates on the creditworthiness of debtor countries

borrowing at these rates. This is a further channel through which low world

interest rates may induce capital to flow into emerging markets.

15See, for example, Calvo et al. (1993), Calvo et al. (1996), Diaz-Alejandro (1983),Fernandez-Arias (1996), and Montiel and Reinhart (1999).

16In addition, Calvo et al. (1993) mention introduction of institutional reforms such asliberalization of the domestic capital market, opening of the trade account, and policies thatresult in credible increases in the rate of return on investment.

17As stated in Calvo et al. (1996), additional external factors include terms-of-trade de-velopments, the international business cycle, and regulatory changes that affect the interna-tional diversification of investment portfolios at the main financial centers.


The relevant literature disputes whether external or internal factors are

more important in determining the direction and composition of these flows.

In the present model it is indeterminate whether the derivative of the threshold

with respect to α is smaller or bigger than the derivative with respect to r in

absolute values.18 So the model does not reveal, which factors, external or

internal, are more important. However, the present model can illustrate that

the scope of government policies dealing with possible coordination failures

changes as a function of external factors. If the international interest rate

increases, governments of developing countries lose scope, whereas they gain

scope if the interest rate falls. This study finds that the government can buy

scope for its policies by, for example, productivity enhancing reforms. But, at

the same time, governments lose scope if the productivity is decreased. This

means that the relative importance of internal versus external factors varies

over time. As governments lose scope when the international interest rate

increases, i.e., when the economic surrounding turns unfavorable, governments

are even in a worse position because their effort in preventing a crisis would

even have less effect.

2.5.3 Changes in the Degree of Uncertainty ε

Finally, the comparative static analysis with respect to the precision of private

information ε reveals interesting insights.

Proposition 2.6 If the degree of uncertainty about the fundamentals ε in-

creases, the threshold equilibrium is shifted to a lower level of debt, i.e., a

growth collapse and, thereby, a sudden stop occurs already at lower levels of

debt.

Proof. Proving 2.6 requires calculating the derivative of the threshold D? with

respect to the variance of the private signal around the true value of debt.

∂D?

∂ε= − z

6(r − 12z)

< 0 (2.20)

Recall that r > z. Hence, the derivative is always negative. This means that

D? decreases with increasing uncertainty.

18This depends on the relative magnitudes of α, z, and r.

2.6. TESTABLE HYPOTHESES 61

As argued before, this finding implies that the probability of a sudden stop

increases with an increase in uncertainty. Formulated differently, this means

that the more precise the information, the lower the probability of a bad equi-

librium. This result contrasts the findings of the ”game of refinancing.”19 In

terms of government policies, these results mean that governments should aim

for an information dissemination policy that entails small variation in the value

of private signals, i.e. that entails little uncertainty about the fundamentals.20

2.6 Testable Hypotheses

In this section, predictions of the theoretical model are translated into a set of

testable hypotheses. In particular, I formulate testable hypotheses regarding

the influence of technological progress, the international interest rate, and the

uncertainty about the fundamentals on the probability of a sudden stop.

Hypothesis 1 Sudden stops become less likely if internal factors of emerging

market countries become more favorable, e.g., if the investment safety increases

or if governments adopt technology enhancing policies (derived from Proposi-

tion 2.2).

Hypothesis 2 Sudden stops become more likely if the international interest

rate increases (see Proposition 2.18).

Hypothesis 3 Sudden stops become more probable with more uncertainty on

the fundamentals of the economy (see Proposition 2.6).

2.7 Empirical Analysis

The purpose of this section is to validate the predictions of the theoretical

model, and evaluate these three hypotheses. Particular attention will be paid

to showing the effect of uncertainty about the fundamentals on the occurrence

of sudden stops of capital flows.

19See Morris and Shin (2004).20For an extensive analysis of transparency, see Heinemann and Illing (2002).


2.7.1 The Data

I work with two data sets: a yearly data set of 14 emerging and 17 industri-

alized countries and a monthly data set of 11 emerging and 14 industrialized

countries. Both sets run from January 1990 to December 2001. I work with

these two data sets because the yearly data does not allow me to tackle the

potential problem of endogeneity in a convincing manner because there are

too few observations. However, the results of the analysis with yearly data

are displayed, because a part of the time series of control variables are only

available in yearly frequency. In the analysis with monthly data, those series

are interpolated.

The selection of countries reflects those emerging countries tracked by JP

Morgan’s Emerging Market Outlook, i.e., countries that significantly show in

the world capital markets. The developed countries in the sample are OECD

members. However, some of the countries that fulfill those criteria are dropped

due to lack of the relevant data.21

The dependent variable is an index of sudden stops of capital flows. Fol-

lowing Calvo et al. (2004), I employ a dummy variable based on monthly data

of capital flows. This high frequency of data is chosen because it best unveils

the origin of sudden stop crisis episodes. Due to the high frequency of data,

however, it is necessary to work with a proxy for the flows, netting out the

trade balance from changes in foreign reserves. Then, the change in the capital

flows with respect to the capital flows 12 months before is calculated to avoid

seasonal effects.

The first criterion that determines whether a month is counted as a crisis

month or not concerns capital flows: This criterion is fulfilled for an observa-

tion where the year-to-year decrease in capital flows lies at least two standard

deviations below its sample mean. To introduce persistence in this measure,

the criterion is also regarded as fulfilled if the flows decrease more than one

standard deviation below the sample mean in the months that encircle the

two standard deviation decrease. In addition to this first criterion, the second

criterion is that the output of the economy has to contract at the same time.

Thereby, only crisis episodes with costly disruptions in economic activity are

identified. For robustness checks, I make also use of a crisis dummy variable,

21For more details, see Table 2.1 in Appendix 2.10.

2.7. EMPIRICAL ANALYSIS 63

for which only the capital flow criterion has to be fulfilled.22 In the analysis

with yearly data, a year is counted as a sudden stop year if it contains at least

one month that fulfils the above-mentioned criteria.

The explanatory variable most interesting for the present analysis is un-

certainty about the fundamentals. I use the standard deviation of growth

forecasts by a group of country experts as a measure of uncertainty. In the

models a la Morris and Shin (1998), uncertainty takes the form of the disper-

sion of the private signals around the true value of the fundamentals. In this

chapter of the study, this translates into the dispersion of the private signals

about the true value of government debt, i.e., the evaluation of the debt level

by each of the private investors. Direct observation of the private signals of all

investors is not possible. However, there is data that can be used as reliable

proxy.

Firstly, given the distributional assumptions that I made in the theoretical

model the expectation of the true value of the debt, given the private signal,

is exactly the private signal itself: E(D|Di) = Di. If the private signals are

dispersed with a standard deviation of ε around the true value of the debt, the

expectations will as well. Therefore, the standard deviation of the expectations

will give a good indication of the standard deviation of the signals of concern

in this chapter. Data are available that closely proxies the expectations by

private agents about the fundamentals.

Data on the standard deviation of expectations about the level of gov-

ernment debt are not available in a sufficient coverage. Therefore, the best

available option is to use the standard deviation of expectations about GDP

growth as a proxy. In doing this, I follow Prati and Sbracia (2002) who use

these data to test the effect of uncertainty on the occurrence of currency crises

22No consensus in the literature exists about the concept of capital flows or the criteria todetect a sudden stop. While, for example, Calvo and Reinhart (2000) examine variations innet private capital flows, Milesi-Ferretti and Razin (1998), and Hutchison and Noy (2006)analyze changes in the current account. In addition, measures of the variation in capitalflows differ. In one part of the literature, negative differences are measured relative tothe country’s GDP and considered a sudden stop if they exceed a specific threshold (see,for example, Calvo and Reinhart (2000) and Hutchison and Noy (2006)). However, newerliterature also takes into consideration the unexpected character of such an extreme eventand considers a drop in capital flows a crisis when it falls below a threshold in terms of thestandard deviations below the sample mean (see Calvo et al. (2004), Cavallo and Frankel(2004), and Eichengreen et al. (2006)). I use the latter approach because it is consistentwith my theoretical model.


in a similar model. In addition, because the model in this chapter of the study

also works if the uncertainty lies on the productivity parameter, this procedure

seems even more justified.

A second restriction is that no data exists on the private signals of all

investors. Data collecting firms only survey the opinions of a group of about 20

banks and other market analysts per country. However, assuming that private

agents can buy expert opinion, it is reasonable that they will buy different

numbers of those opinions and will weigh these signals differently. If the experts

strongly diverge in their expectations, private agents will most likely have

even more divergent evaluations of the fundamentals. Therefore, dispersion of

expert opinions, i.e., their standard deviation, seems a good indicator of the

dispersion of private agents’ expectations about the fundamentals.

I use data from two sources: the IFO Institute for Economic Research and

Consensus Economics. Both institutes collect GDP forecasts of a group of

experts within all the countries that they track at a particular point in time.

The IFO Institute and Consensus Economics then report mean and standard

deviation of these forecasts for the respective country. I use the standard de-

viations as the measure of uncertainty. When working with yearly data, I use

both data sets. While the IFO Institute asks experts within tracked countries

about their forecasts of GDP growth for the current year, once yearly in April

Consensus Economics collects forecasts of GDP growth, CPI inflation, gov-

ernment budget balance, current account balance trade balance, and exports

for the current and following year in monthly frequency. In the analysis with

yearly data I display two sets of estimations: In the first set of estimations, the

measure of uncertainty is a yearly average of the standard deviations of fore-

casts that Consensus Economics gathers. In the second set of estimations, I

combine the observations by Consensus and WES (World Economic Survey by

the IFO Institute). I do this by only taking the April forecast for the current

year by Consensus. If both observations are available, I use the WES data.23

To achieve a constant one-year forecast horizon for the data by Consensus

Economics, I follow Prati and Sbracia (2002) in computing a weighted average

of the current and the following year forecast. In January a weight of 11/12

is attributed to the current and of 1/12 to the following year forecast. In

23For robustness checks, I ran the same regressions two more times: first, using WES dataonly and second, using a combination where in case of redundancy I took the Consensusdata. The results are qualitatively the same and quantitatively similar.


February the weights equal 10/12 and 2/12 respectively. For every month in

the same logic, another set of weights is applicable until December, where the

respective weights are 0/12 and 12/12.24

I use a large set of control variables. Firstly, I control for the mean of the

growths forecasts. This appears to be the most important control because I

want to disentangle the effect of private investors having diverging opinions as

opposed to all investors being sure that growth will be low. Additionally, I

draw upon Calvo et al. (2004). They convincingly put forward the vulnerability

to large real exchange-rate fluctuations and the degree of domestic liability

dollarization as drivers of the occurrence of sudden stops. In addition, I use a

large set of macroeconomic controls. When I work with monthly data, I have

to interpolate several time series of the control variables that are only available

in yearly frequency. This technique does have a cost: an understating of the

variance of those series. Since I have monthly observations on the variables

that I am most interested in and most of the controls that I have to interpolate

represent economic variables that do not vary substantially in a year’s time,

it is very unlikely that this procedure influences the results. In addition, I

can show the presence of the effect of uncertainty on the occurrence of sudden

stops of capital flows with yearly data.

2.7.2 Benchmark Regression

As a benchmark regression I estimate a pooled probit controlling for country

and time effects. The theoretical model is static and predicts the probability

of a crisis at a particular point in time. Therefore, a probit approach to

estimating the effect of uncertainty on the occurrence of a crisis appears most

appropriate.

Prob(Suddenstop = 1|xitβ) = G(β0 + β1unci,t−1

+ β2mgexpi,t−1 + β3macrocntrlsi,t−1

+ δi + γt + εi,t) (2.21)

with i = 1, 2, ..., n; t = 1, 2, ..., T .

24As a robustness check I rerun all the estimations with the current year and with followingyear forecasts separately. The results are qualitatively the same and quantitatively similar.


Suddenstop equals one if country i experiences a sudden stop in period

t. xit represents the set of explanatory variables, including the measure of

uncertainty, explicitly named unc in the right hand side of Equation (2.21), the

mean of the growth expectations, named mgexp, and a set of macroeconomic

control variables, named macrocntrols. For a full list of the variables used

as control variables, please refer to Table 2.2 in Appendix 2.10. δi stands

for country fixed effects, γt stands for time fixed effects, and εi,t represents

the error term. β is the vector of the corresponding coefficients. G(.) is the

standard normal cumulative distribution function.

Country dummies are included into the analysis. The level of uncertainty

varies strongly across countries.25 While in some countries, for example, the

Netherlands or Italy, the average of the uncertainty measure, i.e., the standard

deviation between the growth forecasts of different experts (January 1990 - De-

cember 2001), is as low as 0.24 percentage points, in countries like Indonesia or

Turkey, the average of the uncertainty measure reaches levels of 1.022 and 1.15

respectively. These statistics suggest that, systematically, some countries are

characterized by higher uncertainty than other countries, therefore requiring

control for country fixed effects. In a probit estimation obtaining consistent

estimates requires controlling for time fixed effects by incorporating country

dummies in a pooled regression.

Additionally, I control for time fixed effects. Calvo et al. (2004), in line with

a large part of the literature, state that sudden stops in emerging economies

”bunch” around the Tequila (1994), East Asian (1997), and Russian (1998)

crises. In developed countries, sudden stops materialize mostly around the

ERM (Exchange Rate Mechanism) crisis in 1993. The graphs in Figure 2.3

in the Appendix, which depict the sudden stop periods against the measure

of uncertainty, also show this feature of the crises. Thus, controlling for time

fixed effects is necessary. Mostly I do this by including time dummies in

the regressions. However, where the data quality does not allow for this, I

use polynomial time trends to reduce the number of dummy variables and

approximate the variation over time.

25This is illustrated in Table 2.3 in the Appendix.


2.7.3 Analysis with Yearly Data

I run a pooled probit regression with a sample of all countries with yearly data.

As seen in Table 2.4 in Appendix 2.10, the coefficient on the contemporaneous

uncertainty has a positive sign, irrespective of the measure of uncertainty cho-

sen here. Controlling for country and time effects, the result is significant.26

However, sudden stops are mainly an emerging market phenomenon.27 For

some of the countries that do not experience sudden stops of capital flows, the

country dummies are dropped from the regression, while the observations are

included. This makes the result look weaker. To circumvent this difficulty, the

analysis is repeated for emerging economies only. The results are reported in

Table 2.5 in Appendix 2.10. Again, the effect of the uncertainty on the sudden

stop crisis probability is positive and significant. However, if working with the

uncertainty measure based on the Consensus data in the case of the inclusion

of the quadratic time trend (column 3 in the left half of the table), none of

the explanatory variables is significant. This seems to be related to the small

number of 64 observations. This does not happen with the combined measure

of uncertainty, for which it is possible to obtain 97 observations.

In all regressions, I control for the mean of the expectations over all the

experts. Therewith, I ensure to disentangle the self-fulfilling effect of the ex-

pectations from the uncertainty about the fundamentals, i.e., the disagreement

on the state of the economy. In most of the regressions, the mean of the expec-

tations turns out to significantly impact the crisis probability: The lower the

mean of the expectations, the higher the crisis probability. The other control

variables, namely the domestic liability dollarization, the vulnerability to real

exchange rate fluctuations, the index of exchange rate flexibility, the reserves

over the current account deficit, M2 over reserves, credit growth, foreign di-

rect investment over GDP, public balance over GDP, total debt over GDP,

and TOT growths turn out to be insignificant in many of the regressions when

controlling for country fixed effects.28

This analysis with yearly data suggests that the empirical findings are in

26The results stay the same when including higher order time trend. However, if countryand year dummies are included, none of the explanatory variables are significant, whichindicates that including all these dummy variables is demanding too much from the data.

27See Figure 2.3 in the Appendix.28I present the results with the most convincing specification in terms of the control

variables. However, I have run all the regressions with a larger set of controls and theresults are qualitatively the same and quantitatively similar.


line with the theoretical model. However, due to data limitations, this analysis

does not permit tackling one obvious problem of the analysis: the direction of

causality. Here, the monthly data contributes to finding a remedy.

2.7.4 Analysis with Monthly Data

When repeating the analysis with monthly data, the one-month lag of the ex-

planatory variables is used as a first step to reduce the problem of endogeneity.

Additionally, one month is an appropriate time, on average, that investors can

act according to their expectations. The capital flow proxy comprises portfolio

investments, which are very liquid, and foreign direct investments, which are

less liquid. The regression, which includes the entire country sample, suffers

from the same difficulty of dropped country dummies as the counterpart with

yearly data. The results of these regressions are displayed in Table 2.6. Still,

the analysis shows that the sign of the effect of the uncertainty on the oc-

currence of crises is positive as expected. The pooled probit estimation with

monthly data and the emerging market sample reveals the most interesting

insights into the effect of uncertainty.

(Table 2.7 here)

The results of the analysis with monthly data and the emerging market

sample turn out as expected by theory.29 The lagged uncertainty influences

the crisis probability positively. The lagged expectations themselves have the

opposite impact. The fact that the vulnerability against real exchange rate

fluctuations and the domestic liability dollarization are insignificant may lie

in the interpolation of these series from yearly data. These are the variables

that Calvo et al. (2004) put forward as main drivers of sudden stops. However,

recall that in the analysis with yearly data30, the two variables do not appear

29The result with respect to the uncertainty holds also when applying monthly time dum-mies into the analysis. However, most of the other variables prove insignificant, which mayreflect that some of them are interpolated from yearly data. Therefore, the specificationwith time dummies is not displayed here and is not the most convincing specification.

30See Appendix 2.10, Tables 2.4 and 2.5.


significant in a number of specifications. This finding casts a doubt on the

robustness of the results by Calvo et al. (2004), once country fixed effects are

controlled for. As argued before, a pooled probit analysis using country and

time dummies is a more convincing approach for the current question than the

assumption of random effects as used in Calvo et al. (2004).31

The effect of uncertainty on the occurrence of sudden stops is not negligible.

Assessing the relevance of the effect of uncertainty requires calculation of the

marginal effects for the regressions from Table 2.7. The effect ranges between

a 2.5 and 10.8 percent increase of the crisis probability, if uncertainty (i.e.,

the standard deviation of the growth forecasts) is increased by one percentage

point. The most convincing specifications are those with a quadratic or cubic

time trend. Therefore, it is safe to say that an effect of 2.5 to 5.6 percent is

most realistic. The variation stems from different specifications.

(Table 2.8 here)

2.7.5 Facing Endogeneity

To dispel the possibility that the results presented above stem from an endo-

geneity problem rather than displaying the effect of the uncertainty about the

fundamentals, I first apply higher order lags as explanatory variables. Second,

I implement instrumental variable estimation.

The analysis with lags of the potentially endogenous variables (namely the

uncertainty measure, the mean of the expectations, and the vulnerability to

real exchange rate fluctuations) reveals that the uncertainty up to four month

previous to the crisis period has a significant positive effect on the probability of

a crisis. Earlier uncertainty, however, does not matter for a crisis to occur. This

pattern does not materialize with respect to the mean of the expectations and

the vulnerability to real exchange rate fluctuations, which both stay significant

31The result of the uncertainty influencing the crisis probability positively also holds underthe assumption of random effects controlling for time effects.


when applying these higher order lags. These results are illustrated in Table

2.9 in Appendix 2.10.

It is hard to confirm that a lag of four month is enough to deny possible

endogeneity. Nevertheless, four month prior to a crisis, it is mostly impossi-

ble to predict when and in which form the crisis will materialize. Therefore,

investors cannot be certain that a crisis is going to happen. Furthermore, it

is also not surprising that uncertainty does not have an effect for more than

four month into the future. If the disagreement between investors about future

outcomes at the point of the investment decision itself really matters, then one

month should be a good proxy for the reaction time. To summarize: these

results cannot exclude the possibility that the result is driven by endogeneity,

but these results render it much less likely.

The next step in the attempt to cope with the potential endogeneity is

to instrument the contemporaneous variable with its own lag. In the first

stage, the uncertainty measure is regressed on its own lag controlling for the

same set of controls as in the second stage regression. Based on the estimated

coefficients, the contemporaneous uncertainty is then predicted. In the second

stage, the predicted values are used along with the control variables.

The results in Table 2.10 in Appendix 2.10 suggest that uncertainty does

have an aggravating effect on the probability of a crisis. Here, I display the

results where I used the six-month lags of the potentially endogenous variables

as instruments. I did the same analysis with lower- and higher-order lags.

The results are similar for lower-order lags. For higher-order lags, however,

they break down: Under some specifications, the lags are not significant in the

first-stage regressions any more; and, under other specifications, the predicted

values do not significantly explain the occurrence of a crisis. Keeping in mind

the fast burst speed of many sudden stop events, it seems that a lag of six

months is a sufficient distance to exclude the causality from the crisis to the

uncertainty. In addition (and this also applies already to the argument when

explaining the results with the lagged explanatory variables), I control for

mean expectations and for time effects, which ensures against picking up the

reverse causality by the uncertainty variable. Hence, both the analysis with

lagged explanatory variables and the instrumental variable estimation further

support the validity of the theoretical findings.

One additional possibility to avoid the endogeneity problem would be to

look for past data revisions as an instrument for the uncertainty. However,


there might be also a problem that revisions of data often happen in sight of

a crisis, e.g., to smooth outcomes.

2.7.6 Robustness Analysis

One additional set of estimations is reported in this analysis to check that

the results are not sensitive to the econometric method chosen. Table 2.11

in Appendix 2.10 reports the results of these estimations. Additionally, to

the pooled probit that was chosen as a benchmark case, I estimate a pooled

logit controlling for country and time effects, a conditional logit in a panel

setting with fixed effects, and a Chamberlain’s panel probit estimations. All

these approaches have in common that they control for country specific effects.

Applying the logit estimation implies employing the logistic function instead

of the normal cumulative distribution function as in the probit approach. The

conditional logit allows for a fixed effects estimation, which is not possible

in a probit setting. A fixed effects estimation in a probit setting leads to

inconsistent coefficient estimates because the country effects cannot cancel out

when they are within the cumulative distribution function. The problem is

reduced in the case of the logistic function. Using Chamberlain’s panel probit

approach allows for the unobserved heterogeneity to be correlated with the

mean of each of the explanatory variables, which is calculated by country and

included into the estimation as further control. Therefore, this mean functions

similarly to a country dummy.32

As Table 2.11 in Appendix 2.10 illustrates, the positive effect of uncer-

tainty on the occurrence of sudden stops is robust against different estimation

approaches; the negative effect of the expectations is robust as well.

All regressions are also run with the complete list of control variables.33

The results do not qualitatively and quantitatively change when including the

additional variables. The specification that I show here contains the most

important explanatory variables. The different results illustrate different spec-

ifications in terms of the control for time effects (including time trends of

differing order). This approach is the best available option because some of

the series of control variables are interpolated and, therefore, data quality does

not always allow inclusion of the 144 monthly time dummies in the regressions.

32See Wooldridge (2002), pp. 487f., for a detailed description.33See Table 2.2 in Appendix 2.10.


Furthermore, I run all the above regressions with an alternative measure of

sudden stops. Specifically, the analysis is repeated counting a month as sudden

stop month if the criterion regarding the drop in capital flows is fulfilled while

ignoring whether growth is positive or negative in the respective period. The

results from this analysis are quantitatively the same as the ones that I report

here.

The empirical findings conclude by noting the contemporaneous positive

and significant effect of uncertainty on the crisis probability shown in the

analysis with yearly data. However, recall the difficulty of resolving potential

endogeneity with this data set. One way out of the problem of potential

endogeneity is to switch to monthly data: I first apply higher-order lags as

explanatory variables. In a further step, I perform a two-stage estimation with

the lags of the potentially endogenous variables serving as instruments. This

works for lags up to six months. Additionally, I check for different estimation

approaches. In the analysis with monthly data, the positive and significant

effect of uncertainty persists. By calculating the marginal effects, one can also

show that the effect that I am showing is not negligible quantitatively.

Summarizing, the empirical results support the theoretical predictions of

the model. Recall, the model predicts that the probability of a sudden stop

increases with a deterioration of internal factors. Moreover, it predicts that

the probability of a sudden stop increases with external conditions turning

unfavorable. Most importantly, the model predicts that the probability of a

sudden stop increases with an increase in uncertainty about the fundamentals.

2.8 Policy Implications

This section discusses the implications of the theoretical and empirical results

for economic policies.

The increase in the technology parameter decreases the probability of a

crisis. This implies that governments should try to enhance technological

progress, rendering their country more attractive for investment. Also, in the

context of the technology parameter, the safety of investment seems crucial so

that investors can realize a high after-tax return on investment. In the same

line of argument, high tax policies seem counterproductive. To summarize, all

steps toward a credible increase in the long term rate of return on investment

2.9. CONCLUSION 73

help prevent a crisis. An interesting implication of this analysis is that, apart

from the direct effect of an increase in the technology parameter, a government

can buy scope for other policies that help private investors coordinate on the

good equilibrium if they increase this parameter.

The international interest rate is not under the control of one economy.

The model, rather, refers to small open economies. The findings in this con-

text imply that governments should take into consideration that they have

even less scope for action once the outside world turns unfavorable, i.e., if

the international interest rate increases. Therefore, governments should take

precautions for such cases of deteriorating external conditions.

Another finding of the present analysis is that private information with

little noise is the most favorable setting for an economy. Therefore, a govern-

ment should achieve such an informational structure to advance their interest

in preventing a sudden stop event.

In the present model, the government mechanically services its debt in a

static model. This means that problems of credibility or commitment are ab-

stracted from. Therefore, one can only infer policy implications for a credible

government. This model does not cover the mechanisms by which govern-

ments could achieve credibility. One venue by which a credible government

could achieve a preferred setting, in which investors decide upon private infor-

mation, would be to allow full access to government data to a small group of

independent economic rating agencies. These institutions would be allowed to

gather all relevant information on the fundamentals and could then sell their

signals to private investors. The private investors could buy signals of different

agencies, weighing those according to their preferences. This would make sure

that signals, which the investors in the market have about the fundamentals,

would be private and characterized by a small amount of noise.

2.9 Conclusion

This study considers the possibility of coordination failure between investors

as a factor triggering a sudden stop. This finding is verified empirically. More

specifically, the analysis shows that an increased uncertainty about the fun-

damentals of an economy increases the probability of a sudden stop of capital

flows.


The main theoretical findings of this chapter about the causes of sudden

stop events are the following: 1) The probability of a sudden stop decreases

with technological progress. 2) The probability of a sudden stop increases

with a higher international interest rate. 3) The probability of a sudden stop

also increases with noisier private signals, which can be interpreted as higher

uncertainty or disagreement about the fundamentals among private investors.

With regard to the discussion on internal versus external factors that at-

tract capital to emerging markets, this study finds that, with increasing inter-

national interest rate, the scope of policy action, preventing a sudden stop, is

reduced. In contrast to this result, the government gains scope for its actions

through the use of policies that advance investment safety or technological

progress. Thus, in terms of the discussion regarding pull and push factors of

capital flows, the outcome is that the relative importance of those factors vary

over time in an unfavorable way for the concerned economies. If the external

conditions are unfavorable, governments have less possibility to influence the

economic outcome by, for example, helping private investors to coordinate on

the good equilibrium.

This study contributes to the empirical literature on sudden stops of capital

flows by the verification of the theoretical findings on the effect of uncertainty

on the probability of a sudden stop. To verify the theoretical result, a pooled

probit analysis controlling for country and time effects is used. Calculating

marginal effects also shows that the influence of the uncertainty on the oc-

currence of sudden stops is quantitatively not negligible. Additionally, a large

number of robustness checks is implemented. These include two-stage estima-

tions to address the potential endogeneity problem. In all these regressions,

the positive effect of the uncertainty about the fundamentals on the probability

of the occurrence of a crisis persists.

The results strongly suggest that governments should take uncertainty

about the fundamentals in the economy into account. Lower precision of infor-

mation about the government’s fiscal policy and, therefore, uncertainty about

these values increases the probability of a sudden stop of capital flows. Hence,

an economy will be more vulnerable in times when uncertainty is higher. Pol-

icymakers should adjust their policies accordingly. Specifically, the provision

of less noisy private information is crucial in this context. One venue would

be to allow full access to all government data to a small set of independent

agencies which could then sell their ratings to private investors.

2.9. CONCLUSION 75

Limitations of the present study are that considerations about default and

thereby credit frictions have not been included. Furthermore, the analysis has

not extended the non-monetary model to a monetary one. Calvo (2003) illus-

trates these extensions in his model. In the mentioned paper, Calvo also shows

that foreseen a crisis is possible in the model. The introduction of infinitely

many firms and the coordination problem do not alter these considerations.

Banking crises, however, cannot be rationalized within the current framework.

However, looking at these factors would be an interesting agenda for future

work.

For future research, two potential extensions of this theoretical model would

be interesting. First, adding the assumption of public information about the

debt level to the assumption of private information of each agent could be

worthwhile. So far, this study assumes that agents base their decisions on

personal interpretation of publicly available information, i.e., each agent does

not know how the other agents interpret the available information. Morris and

Shin (2004), Metz (2002), and Hellwig (2002) include public information that is

known by all agents into their analysis. However, it is questionable whether this

would generate different implications in the present setup. A vivid discussion

on the interaction between public and private information exists in the context

of central bank policy, triggered by Morris and Shin (2004).

Second, analyzing the distinction between domestic and foreign investors

tackling the following questions could generate valuable insights. How would

the probability of a sudden stop be influenced if the signals of domestic and

foreign investors are differently dispersed around the true value of the debt?

Are economies with investors that differ with regard to the precision of their

information more prone to a crisis than economies with homogenous and only

domestic investors? Corsetti, Dasgupta, Morris, and Shin (2004) analyze the

effect of the presence of one big investor who is better informed than the rest

on the occurrence of a currency crisis. In their setup, the large, well-informed

speculator makes all other investors more aggressive. In the present setup

if, for example, domestic investors are better informed and can coordinate

more easily among themselves than foreign investors, one would expect all

investors stampeding to the exits more often. Given the progressing financial

integration of emerging market economies, this question appears relevant for

future research.


2.10 Appendix

2.10.1 Value of the Firm

The firm is characterized by a linear production function yit = αK i

t , where yit

is firm i’s output in period t, α is the productivity factor and Kit is the capital

invested.

The cash flow of firm i in period t is

Sit = α(1− τ)K i

t − Kit (2.22)

where τ represents a constant output tax rate and Kt the rate of capital ac-

cumulation, neglecting capital depreciation.

The value of the firm at t = 0 is

V i =

∫ ∞

0

Site−rtdt (2.23)

where r represents the constant international interest rate.

zit, the growth rate of capital, is defined as zi

t =Ki

t

Kit. Normalizing the initial

capital stock to one, solving the differential equation, and assuming that the

firms only choose between constant growth paths, the value of the firm can be

expressed in the following simplified form:

V i = [α(1− τ)− zi]1

(r − zi)

In order to obtain this simple expression, one has to proceed in several

steps. In a first step, one expresses Kit and Ki

t in Equation (2.22) in terms of

the initial capital stock K0:

First, one solves the differential equation zit =

Kit

Kit. This can be expressed as

∫ t

0

zisds =

∫ t

0Ki

s

Kisds

=∫ t

0dKi

s

ds1

Kisds =

∫ t

01

KisdK i

s

= lnKit − lnKi

0

⇔ e∫ t0 zi

sds =Ki

t

Ki0

⇔ K it = e

∫ t0 zi

sdsK i0

2.10. APPENDIX 77

Second, noting that K it = zi

tKit , plugging K i

t and Kit into Equation (2.22), and

normalizing K i0 to unity delivers

Sit = α(1− τ)e

∫ t0 zi

sds − zite

∫ t0 zi

sds

Plugging the new expression of Sit into Equation (2.23) leads to

V i =

∫ ∞

0

[(α(1− τ)e∫ t0 zi

sds − zite

∫ t0 zi

sds)e−rt]dt

Third, because r is constant, e−rt can be expressed as e−∫ t0 rdt. Thus, the above

expression can be rearranged to

⇔ V i =

∫ ∞

0

[α(1− τ)− zit]e

− ∫ t0 (r−zi

s)dsdt

Given that firms choose between constant growth paths in the setup, this

expression can be simplified to

⇔ V i =∫∞0

[α(1− τ)− zi]e−(r−zi)tdt

= [α(1− τ)− zi]∫∞

0e−(r−zi)tdt

and, assuming r>z

= [α(1− τ)− zi][ e−(r−zi)t

−(r−zi)]∞0 = [α(1− τ)− zi]

1

(r − zi)

2.10.2 Lemma Proofs

Proof of Lemma 2.1

Proof. With the switching strategy IT , the fraction of firms investing is

π−i(IT ) =∫ 1

0IT (Dj)dj. Thereby, the expected value in Equation (2.11) is

E(Dπ−i(Dj)|Di) = E(Dπ−i(IT )|Di).

By the use of the law of iterated expectations and the fact that D is more

precise information than the private signals Di and Dj, it is known that34

E(Dπ−i(IT )|Di) = E(E(Dπ−i(IT )|D)|Di)

34In general, it is known that if one has an information set Ω3 = (Ω1,Ω2), then theexpectation of a random variable X conditional on a small information set Ω2, E(X|Ω2)is equivalent to the conditional expectation given this smaller set Ω2 of the conditionalexpectation given the bigger information set Ω3 of X: E(E(X|Ω3)|Ω2).


This is equivalent to

E(DE(π−i(IT )|D)|Di)

As the signals, given D, are independent of each other, the expected fraction

of firms that receive a signal smaller than some threshold T is equal to the

probability that one firm receives such a signal given the realization of D:

E(π−i(IT )|D) = prob(Dj ≤ T |D)

Proof of Lemma 2.2

Proof. The support of the distribution of the true value of the debt [D, D]

exceeds D and D (the borders of the multiplicity area in terms of the true

value of the fundamentals, which were found in section 2.3.1, by at least more

than ε each. Therefore, there exist signals D0 and D0, such that

E(D|D0) = D and E(D|D0) = D

and as

E(D|Di = D) = D and E(D| Di = D) = D

this implies that D0 = D and D0

= D.

If firm i receives a signal of exactly D, and even in the worst case that the

probability of another firm investing was 0, the payoff difference equals 0 given

this signal.35

Proof of Lemma 2.3

Proof. The monotonicity of P in Di is a necessary condition for the iterated

elimination of dominated strategies to work and to make sure that there are

not several values for which Equation (2.14) holds.

The factor z 1(r−z)r

is positive, thus I focus on the rest of the expression. It

is clear that the term −rDi is strictly decreasing in Di:

35Plug the right hand side of Equation (2.6) into Equation (2.13) and set prob(Dj <K|D)|Di = 0).

2.10. APPENDIX 79

∂(−zrDi)

∂Di= −r < 0

.

It is more difficult to show the characteristics of the term E(D∗prob(Dj <

T |D)|Di): Making use of the distributional assumptions that I made with re-

gard to the true value of debt and the private signal, I can write the conditional

density of the private signal, given the true value of the fundamentals, in the

following way:

g(Di|D) =

12ε

if D − ε ≤ Di ≤ D + ε0 otherwhise

(2.24)

Therefore, I can write prob(Dj < T |D) as

prob(Dj < T |D) =

0, if T < D − ε12ε

(T −D + ε), if D − ε ≤ T ≤ D + ε1, if T > D + ε

(2.25)

Now, in addition referring to the conditional density of the true value of

debt given the private signal that firm i receives,

h(D|Di) =

12ε

if Di − ε ≤ D ≤ Di + ε0 otherwhise

(2.26)

One can rewrite the expected value as:

E(Dprob(Dj < T |D)|Di) =

∫ Di+ε

Di−ε

Dprob(Dj < T |D)1

2εdD

=

∫ Di+ε

Di−εD2ε

0dD, if T < Di − 2ε∫ T+ε

Di−εD2ε

12ε

(T −D + ε)dD

+∫ Di+ε

T+εD2ε

0dD, if Di − 2ε ≤ T ≤ Di

∫ T−ε

Di−εD2ε

1dD

+∫ Di+ε

T−εD2ε

12ε

(T −D + ε)dD, if Di ≤ T ≤ Di + 2ε∫ Di+ε

Di−εD2ε∗ 1dD, if Di + 2ε < T

(2.27)

The value of the conditional probability depends on the relative position

of T to D and, therefore, the expectation of it given Di also depends on the


relative position of T to Di. Due to the fact that Di is known to the firm, the

integral is evaluated from Di − ε to Di + ε.

Equation (2.27) delivers:

=

0, if T < Di − 2ε1

4ε2

(13((Di)3)− 1

2(3ε + T )(Di)2

+(2ε2 + Tε)Di + 16T 3 + 1

2T 2ε− 2

3ε3

), if Di − 2ε ≤ T ≤ Di

14ε2

(− 13((Di)3)− 1

2(3ε− T )(Di)2

+(2ε2 + Tε)Di − 16T 3 + 1

2T 2ε− 2

3ε3

), if Di ≤ T ≤ Di + 2ε

Di, if D + 2ε < T

(2.28)

I have shown that the term −rDi is strictly monotonically decreasing in

Di. In addition, one knows that z < r. Hence, for the monotonicity of the

expected payoff difference between investing and not investing, P (Di, IT ), it

is sufficient that the derivative of the expected value that I am analyzing is

smaller or equal to 1. The derivatives for the different intervals of the expected

value are the following:

∂E(D ∗ prob(Dj < T |D)|Di)

∂Di=

0, if T < Di − 2ε1

4ε2

((Di)2 − (3ε + T )Di

+(2ε2 + Tε)), if Di − 2ε ≤ T ≤ Di

14ε2

(− (Di)2 − (3ε− T )Di

+(2ε2 + Tε)), if Di ≤ T ≤ Di + 2ε

1, if D + 2ε < T(2.29)

For the cases of T < Di−2∗ε and D+2ε ≤ T , it is clear that the derivatives

are 0 or 1 respectively and hence that P (Di, IT ) is monotonically decreasing

in Di in these intervals.

For the case that Di − 2ε ≤ T ≤ Di, the derivative is a positive quadratic

function (U-shape) in Di. So the derivative will take its maximum value at

either of the borders of the analyzed interval.

Evaluated at T + 2ε, the function

1

4ε2

((Di)2 − (3ε + T )Di + (2ε2 + Tε)

)

takes the value of 0. Evaluated at T , the function takes the value of 12

(1− T

ε

)if

T ≥ −ε. This is the maximum value that the derivative takes in the mentioned

interval. ε is a very small positive number and T is bound to be positive by

the support of D, hence the restriction is not binding.

2.10. APPENDIX 81

So one can conclude that also in the interval of Di − 2ε ≤ T ≤ Di the

expected payoff difference is monotonically decreasing.

If Di ≤ T ≤ Di + 2ε, I find the following results. First of all, the function

1

4ε2

(− (Di)2 − (3ε− T )Di + (2ε2 + Tε))

is a negative quadratic function in Di (inverse U-shape). So it is necessary

to find out how the function looks in the relevant interval, especially, whether

the maximum of the function lies within it. This can be analyzed by taking

the second derivative of the expected value. If it is positive over the entire

interval, one knows that the analyzed interval is entirely located on the in-

creasing branch of the function. Hence the function takes the maximum value

at the upper limit of the interval. Accordingly, for an entirely negative second

derivative, the interval lies in the decreasing branch and the function will take

its maximum value at the lower limit of the interval. If the second derivative

changes sign the situation is more complicated. Then, one has to find the

maximum of the function.

In the present case, the second derivative of the expected value is

∂2E(Dprob(Dj < T |D)|Di)

∂(Di)2=

1

4ε(−2Di − 3ε + T )

This is a linear function in Di. Plugging in the borders of the interval, one

can determine the sign over the interval: At T − 2ε, the function takes the

value 14ε2

(−T + ε) < 0 if T ≥ ε. As the upper bound of the dominance region

where all firms invest, D, is at least ε bigger than D and a signal within the

dominance region cannot be a switching signal, the restriction of T ≥ ε is not

binding.

At T the second derivative takes the value of 14ε2

(−T − 3ε) < 0 if T >

−3ε, where clearly the restriction is not binding. So the interval that one is

interested in is entirely located in the declining branch of the negative quadratic

function of the first derivative of the expected value. Therefore, the maximum

of the first derivative will be at Di = T − 2ε. It is:

1

4ε2

(− (T − 2ε)2 − (3ε− T )(T − 2ε) + (2ε2 + Tε))

= 1

This is sufficient to proof monotonicity. One can conclude that the expected

payoff difference is strictly monotonically decreasing in the private signal Di.


To complete the evaluation of the function at the borders of the interval

and thereby completing the proof of continuity of the first derivatives, I also

show the value of the function at the upper bound T of the interval. Then

1

4ε2

(− (Di)2 − (3ε− T )Di + (2ε2 + Tε))

takes on the value of 12

(1− T

ε

)and 1

2

(1− T

ε

) ≤ 1 if T ≥ −ε. This is the same

value as when I evaluated the lower bound K for the interval Di−2ε ≤ T ≤ Di.

At all borders of intervals, the derivatives coincide; this indicates the continuity

of the first derivatives and shows the smoothness of the expected value. One

can show continuity for the expected value itself as well.

The above expression for the derivative of the expected value can hence be

expressed as follows:

∂E(Dprob(Dj < T |D)|Di)

∂Di=

0, if T < Di − 2ε∈ [

0, 12(1− T

ε)], if Di − 2ε ≤ T ≤ Di

∈ [12(1− T

ε), 1

], if Di ≤ T ≤ Di + 2ε

1, if D + 2ε ≤ T(2.30)

Adding the two terms that are dependent on Di, one finds that P (Di, IT )

is strictly monotonically decreasing in Di.

Proof of Lemma 2.4

Proof. Finding the solution to Equation (2.14) implies setting T = Di = D?

in Equation (2.27). It is straight forward that I do not have to take into

consideration the cases where T < Di − 2ε and T > Di + 2ε. From Equation

(2.27), I see that for K = Di = D? the second term in the case of Di − 2ε ≤K ≤ Di disappears and the first term becomes

∫ D?+ε

D?−ε

D

2ε

D? −D + ε

2εdD (2.31)

In the case of Di ≤ T ≤ Di + 2ε, the first term disappears and the second

term is identical with expression (2.31).

Solving the integral delivers:

=1

4ε2

[1

2D2D? − 1

3D3 +

1

2εD2

]D?+ε

D?−ε

2.10. APPENDIX 83

The above expression simplifies to:

D?

2− ε

6

With this result, I can simplify Equation (2.14) to become

P (D?) = zi α− r − rD? + z(D?

2− ε

6)

(r − zi)r= 0

Solving for D? delivers the unique value

D? =α− r − ε

6z

(r − 12z)

2.10.3 Iterated Elimination of Dominated Strategies

One starts the elimination at the borders of the multiplicity area.

Due to the strict monotonicity in Di, there exist unambiguous signals D1

<

D0

= D and D1 > D0 = D, such that:

U(Di, ID

0) < 0 for all Di > D1

and U(Di, ID0) > 0 for all Di < D1

As D0

> D0, it also holds that D1

> D1. For the case of the upper border

of the multiplicity area, this means: Given that the other firms do not invest

when receiving signals above D0, the investment does not pay for signals above

D1

either. One finds D1

by calculating U(Di = D1, I

D0). This process can

be iterated. Given that the other firms do not invest when receiving signals

above Dn, it does not pay to invest at a signal D

n+1with D

n+1< D

n. The

signals Dn+1

are found by setting the expected payoff difference to 0, reflecting

indifference between investment and no investment at firm i:

U(Dn+1

, IDn) = zi α− r − rD

n+1+ zE(Dprob(Dj < D

n|D)|Di = Dn+1

)

(r − zi)r= 0

(2.32)

The sequence Dn

is decreasing, monotone and bounded. By the common

knowledge of rationality, this process is driven to its limit of D?

= limn→∞Dn.

Concretely, one finds a value D?

such that

U(D?, ID

?) = 0 (2.33)


D?

has the interpretation that above this signal all firms do not invest with

certainty.

At the lower bound of the multiplicity area, the analogue situation occurs,

just with the sequence Dn being increasing. There one iterates until one finds:

U(D?, ID?) = 0 (2.34)

.

This means, one iterates until one finds a maximum [minimum] signal at

which firm i is indifferent between investing and not, and which is at the same

time the threshold of the switching strategy of all other firms, when starting

off at D0

= D [D0 = D].

The switching strategies ID? and ID? are Nash equilibria of the private

information game. Milgrom and Roberts (1990) have shown that that in all

games with strategic complementarity the set of strategies that resist the iter-

ative elimination of dominated strategies are limited by Nash equilibria. Nash

equilibria are not eliminated through this process. Therefore, ID? and ID? are

the limiting Nash equilibria of the game. If D?

= D?, there exists an unam-

biguous signal D?, below which in equilibrium all firms will invest and above

which no one does.

2.10. APPENDIX 85

2.10.4 Figures and Tables

Arg

enti

na, B

razi

l, C

hile

, Col

ombi

a, C

zech

R

epub

lic, I

ndon

esia

, Sou

th K

orea

, Mex

ico,

Per

u,

Tha

iland

, Tur

key

Aus

tral

ia, C

anad

a, S

witz

erla

nd, G

erm

any,

S

pain

, Fra

nce,

Uni

ted

Kin

gdom

, Ita

ly, J

apan

, N

ethe

rlan

ds, N

orw

ay, N

ew Z

eala

nd,

Sw

eden

, US

A

Con

sens

us E

cono

mic

s Fo

reca

sts

(Con

sens

us E

cono

mic

s)

Arg

enti

na, B

razi

l, C

hile

, Col

umbi

a, C

zech

R

epub

lic, E

cuad

or, I

ndon

esia

, Sou

th K

orea

, M

exic

o, P

eru,

Phi

lippi

nes,

Tha

iland

, Tur

key,

Sou

th

Afr

ica

Aus

tral

ia, C

anad

a, S

witz

erla

nd, G

erm

any,

D

enm

ark,

Spa

in, F

inla

nd, F

ranc

e, U

nite

d K

ingd

om, I

taly

, Jap

an, N

ethe

rlan

ds, N

orw

ay,

New

Zea

land

, Por

tuga

l, S

wed

en, U

SA

Wor

ld E

cono

mic

Sur

vey

(IFO

In

stitu

te)

Em

ergi

ng M

arke

ts C

ount

ries

Indu

stri

aliz

ed C

ount

ries

Arg

enti

na, B

razi

l, C

hile

, Col

ombi

a, C

zech

R

epub

lic, I

ndon

esia

, Sou

th K

orea

, Mex

ico,

Per

u,

Tha

iland

, Tur

key

Aus

tral

ia, C

anad

a, S

witz

erla

nd, G

erm

any,

S

pain

, Fra

nce,

Uni

ted

Kin

gdom

, Ita

ly, J

apan

, N

ethe

rlan

ds, N

orw

ay, N

ew Z

eala

nd,

Sw

eden

, US

A

Con

sens

us E

cono

mic

s Fo

reca

sts

(Con

sens

us E

cono

mic

s)

Arg

enti

na, B

razi

l, C

hile

, Col

umbi

a, C

zech

R

epub

lic, E

cuad

or, I

ndon

esia

, Sou

th K

orea

, M

exic

o, P

eru,

Phi

lippi

nes,

Tha

iland

, Tur

key,

Sou

th

Afr

ica

Aus

tral

ia, C

anad

a, S

witz

erla

nd, G

erm

any,

D

enm

ark,

Spa

in, F

inla

nd, F

ranc

e, U

nite

d K

ingd

om, I

taly

, Jap

an, N

ethe

rlan

ds, N

orw

ay,

New

Zea

land

, Por

tuga

l, S

wed

en, U

SA

Wor

ld E

cono

mic

Sur

vey

(IFO

In

stitu

te)

Em

ergi

ng M

arke

ts C

ount

ries

Indu

stri

aliz

ed C

ount

ries

Table 2.1: Country samples


Var

iabl

es

Exp

ecte

d ef

fect

on

cri

sis

prob

abili

tyM

easu

res

Sour

ces

Dep

ende

nt V

aria

ble:

C

apit

al f

low

pro

xy: M

onth

ly d

ata

on tr

ade

bala

nce

min

us c

hang

es in

inte

rnat

iona

l res

erve

s.

Eva

luat

ed in

199

5 U

S do

llar

sC

alvo

, Izq

uier

do, M

ejia

(20

04),

C

IM(2

004)

: IM

F: I

FSY

earl

y G

row

th r

ate

of G

ross

Dom

esti

c Pr

oduc

t C

IM (

2004

): I

MF:

IFS

Exp

lana

tory

Var

iabl

e:

Unc

erta

inty

mea

sure

posi

tive

Mon

thly

dat

a of

the

wei

gthe

d av

erag

e of

the

stan

dard

dev

iati

ons

of th

e cu

rren

t and

fol

low

ing

year

fo

reca

st o

f ec

omic

gro

wth

. The

sta

ndar

d de

viat

ion

is c

alcu

late

d fo

r th

e ex

pect

atio

ns th

at e

xper

ts

for

the

spec

ific

eco

nom

y ut

ter

at a

par

ticu

lar

poin

t in

tim

e. I

n Ja

nuar

y th

e st

anda

rd d

evia

tion

of

the

curr

ent y

ear

fore

cast

s is

wei

ghte

d w

ith

11/1

2 an

d th

e st

anda

rd d

evia

tion

of

the

foll

owin

g ye

ar

fore

cast

s w

ith

1/12

. In

Febr

uary

the

curr

ent y

ear

valu

e re

ceiv

es 1

0/12

wei

ght a

nd th

e fo

llow

ing

year

one

2/1

2. T

his

sche

me

cont

inue

s un

til D

ecem

ber

wit

h a

wei

ghti

ng o

f 0/

12 f

or th

e cu

rren

t ye

ar v

alue

and

12/

12 f

or th

e fo

llow

ing

year

val

ue. W

hen

wor

king

wit

h ye

arly

dat

a, w

e bu

ild

the

mea

n of

the

12 m

onth

ly v

alue

s.

Con

sens

us E

cono

mic

s

Con

trol

Var

iabl

es:

Vul

nera

bili

ty to

rea

l exc

hang

e ra

te f

luct

uati

ons

posi

tive

As

show

n in

Cal

vo, I

zqui

erdo

, Mej

ia (

2004

), th

e fr

acti

on o

f th

e cu

rren

t acc

ount

def

icit

e re

lati

ve to

th

e de

man

d fo

r tr

adab

le g

oods

in a

n ec

onom

y is

a g

ood

indi

cato

r fo

r th

e vu

lner

abil

ity

agai

nst r

eal

exch

ange

rat

e fl

uctu

atio

ns. T

his

mea

sure

can

be

inte

rpre

ted

as th

e pe

rcen

tage

fal

l in

the

dem

and

for

trad

able

s ne

eded

to c

lose

the

curr

ent a

ccou

nt g

ap.

CIM

(200

4)

The

sum

of

agri

cult

ural

and

indu

stri

al o

utpu

t min

us e

xpor

ts is

use

d to

pro

xy im

port

s an

d th

e pa

rt

of tr

adab

le o

utpu

t tha

t is

cons

umed

dom

esti

call

y. T

he s

hare

of

this

trad

able

out

put i

s bu

ilt r

elat

ive

to to

tal G

DP

at c

onst

ant p

rice

s. T

hen

the

shar

e of

trad

ale

outp

ut in

tota

l out

put i

s m

ulti

plie

d w

ith

the

tota

l dol

lar

GD

P.

Wor

ldba

nk: W

DI

Cur

rent

acc

ount

def

icit

IMF:

WE

O

Dom

esti

c L

iabi

lity

Dol

lari

zati

onpo

siti

veD

evel

oped

cou

ntri

es: L

ocal

ass

et p

osit

ions

in f

orei

gn c

urre

ncy

by B

IS r

epor

ting

ban

ks a

s a

shar

e of

GD

P. E

Ms:

Dol

lar

depo

sits

plu

s ba

nk f

orei

gn b

orro

win

g as

a s

hare

of

GD

P.

CIM

(20

04):

BIS

, Hon

ohan

and

Shi

(20

02),

C

entr

al B

anks

of

Aus

tral

ia, N

ew Z

eala

nd,

Col

umbi

a, K

orea

, Bra

zil,

IMF:

IFS

TO

T g

row

thne

gati

veT

erm

s of

trad

e on

goo

ds a

nd s

ervi

ces,

ann

ual r

ate

of c

hang

eC

IM (

2004

): W

orld

bank

: WE

OT

otal

Deb

t ove

r R

even

ues

posi

tive

CIM

(20

04)

Dev

olop

ed c

ount

ries

: Pub

lic

debt

fro

m O

EC

D. E

Ms:

WD

I; G

ross

cen

tral

gov

ernm

ent d

ebt

OE

CD

, Wor

ldba

nk: W

DI

Rev

enue

s of

the

cent

ral g

over

nmen

tIM

F: W

EO

Res

erve

s ov

er C

AD

nega

tive

CIM

(20

04)

Inte

rnat

iona

l res

erve

sIM

F: I

FSC

urre

nt a

ccou

nt d

efic

itIM

F: W

EO

Ex.

Reg

ime

3 po

siti

veE

xcha

nge

rate

reg

ime

clas

sifi

cati

on in

to 3

cat

egor

ies:

1=

floa

t, 2=

inte

rmed

iate

, 3=

fix

Lev

y-Y

eyat

i and

Stu

rzen

egge

r (2

002)

Ex.

Reg

ime

5 po

siti

veE

xcha

nge

rate

reg

ime

clas

sifi

cati

on in

to 5

cat

egor

ies:

1=

inco

nclu

sive

, 2=

flo

at, 3

= d

irty

, 4=

dirt

y/cr

awli

ng p

eg, 5

= f

ixL

evy-

Yey

ati a

nd S

turz

eneg

ger

(200

2)C

redi

t Gro

wth

nega

tive

Cre

dit t

o pr

ivat

e se

ctor

, ann

ual r

ate

of c

hang

eC

IM (

2004

): I

MF:

IFS

FDI/

GD

Pne

gati

veN

et f

orei

gn d

irec

t inv

estm

ent

CIM

(20

04):

IM

F: I

FSPu

blic

Bal

ance

/GD

P po

siti

veB

alan

ce o

f ge

nera

l gov

ernm

ent

CIM

(20

04):

IM

F: W

EO

M2

over

Res

erve

spo

siti

veM

oney

plu

s qu

asi-

mon

eyC

IM (

2004

): I

MF:

IFS

Sudd

en s

top

indi

cato

r

Table 2.2: List of variables

2.10. APPENDIX 87

coun

try

Mea

n (o

ver

tim

e) o

f m

ean

(ove

r ex

pert

s) o

f gr

owth

exp

ecta

tion

s by

co

untr

y

Max

(ov

er ti

me)

of

mea

n (o

ver

expe

rts)

of

grow

th e

xpec

tati

ons

by

coun

try

Min

(ov

er ti

me)

of

mea

n (o

ver

expe

rts)

of

grow

th

expe

ctat

ions

by

coun

try

Mea

n (o

ver

tim

e) o

f st

anda

rd d

evia

tion

(ov

er

expe

rts)

of

grow

th

expe

ctat

ions

by

coun

try

Max

(ov

er ti

me)

of

stan

dard

dev

iati

on (

over

ex

pert

s) o

f gr

owth

ex

pect

atio

ns b

y co

untr

y

Min

(ov

er ti

me)

of

stan

dard

dev

iati

on (

over

ex

pert

s) o

f gr

owth

ex

pect

atio

ns b

y co

untr

y

Num

ber

of m

onth

s th

at

are

coun

ted

as s

udde

n st

ops

Arg

enti

na3.

296.

45-3

.30

0.87

2.80

0.37

26A

ustr

alia

3.14

4.26

1.14

0.51

0.94

0.24

8B

razi

l2.

694.

67-2

.80

0.76

1.73

0.33

0C

anad

a2.

753.

86-0

.15

0.42

0.70

0.15

0C

hile

5.20

8.93

2.10

0.49

0.80

0.24

12C

olom

bia

3.30

5.30

0.95

0.58

1.12

0.25

12C

zech

Rep

ubli

c2.

765.

280.

250.

530.

930.

283

Fran

ce2.

373.

560.

380.

270.

580.

140

Ger

man

y2.

024.

06-0

.55

0.33

0.64

0.11

5In

done

sia

4.63

7.56

-7.7

11.

023.

320.

1811

Ital

y2.

073.

370.

620.

240.

510.

060

Japa

n1.

714.

91-0

.96

0.67

1.30

0.21

12M

exic

o3.

435.

48-1

.60

0.58

1.17

0.20

7N

ethe

rlan

ds2.

483.

810.

710.

240.

510.

090

New

Zea

land

2.68

4.04

1.49

0.53

0.95

0.30

0N

orw

ay2.

443.

810.

340.

470.

940.

230

Peru

4.07

6.90

-1.3

50.

671.

200.

306

Sout

h K

orea

5.83

7.96

-1.4

00.

822.

110.

2011

Spai

n2.

854.

650.

500.

240.

380.

109

Swed

en1.

762.

840.

700.

260.

600.

1414

Swit

zerl

and

1.76

2.84

0.70

0.26

0.60

0.14

0T

hail

and

4.15

8.53

-3.3

10.

952.

840.

2519

Tur

key

3.23

5.38

-0.9

71.

151.

900.

5926

Uni

ted

Kin

gdom

2.11

3.42

-0.1

70.

420.

850.

190

Uni

ted

Stat

es2.

534.

13-0

.14

0.37

0.76

0.13

12

Table 2.3: Descriptive statistics


Australia - Sudden Stop Dummies and the uncertainty about the fundamentals

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

1990:11 1991:12 1993:01 1994:02 1995:03 1996:04 1997:05 1998:06 1999:07 2000:08 2001:09

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

dum_SS_od s_cfw

Canada - Sudden Stop Dummies and the uncertainty about the fundamentals

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

1990

:01

1990

:08

1991

:03

1991

:10

1992

:05

1992

:12

1993

:07

1994

:02

1994

:09

1995

:04

1995

:11

1996

:06

1997

:01

1997

:08

1998

:03

1998

:10

1999

:05

1999

:12

2000

:07

2001

:02

2001

:09

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

dum_SS_od s_cfw

Switzerland - Sudden Stop Dummies and the uncertainty about the fundamentals

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

1998

:06

1998

:08

1998

:10

1998

:12

1999

:02

1999

:04

1999

:06

1999

:08

1999

:10

1999

:12

2000

:02

2000

:04

2000

:06

2000

:08

2000

:10

2000

:12

2001

:02

2001

:04

2001

:06

2001

:08

2001

:10

2001

:12

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

dum_SS_od s_cfw

Germany - Sudden Stop Dummies and the uncertainty about the fundamentals

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

1990

:01

1990

:08

1991

:03

1991

:10

1992

:05

1992

:12

1993

:07

1994

:02

1994

:09

1995

:04

1995

:11

1996

:06

1997

:01

1997

:08

1998

:03

1998

:10

1999

:05

1999

:12

2000

:07

2001

:02

2001

:09

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

dum_SS_od s_cfw

Spain - Sudden Stop Dummies and the uncertainty about the fundamentals

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

1995

:01

1995

:05

1995

:09

1996

:01

1996

:05

1996

:09

1997

:01

1997

:05

1997

:09

1998

:01

1998

:05

1998

:09

1999

:01

1999

:05

1999

:09

2000

:01

2000

:05

2000

:09

2001

:01

2001

:05

2001

:09

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

dum_SS_od s_cfw

France - Sudden Stop Dummies and the uncertainty about the fundamentals

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

1990

:01

1990

:08

1991

:03

1991

:10

1992

:05

1992

:12

1993

:07

1994

:02

1994

:09

1995

:04

1995

:11

1996

:06

1997

:01

1997

:08

1998

:03

1998

:10

1999

:05

1999

:12

2000

:07

2001

:02

2001

:09

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

dum_SS_od s_cfw

UK - Sudden Stop Dummies and the uncertainty about the fundamentals

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

1990

:01

1990

:08

1991

:03

1991

:10

1992

:05

1992

:12

1993

:07

1994

:02

1994

:09

1995

:04

1995

:11

1996

:06

1997

:01

1997

:08

1998

:03

1998

:10

1999

:05

1999

:12

2000

:07

2001

:02

2001

:09

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

dum_SS_od s_cfw

Italy - Sudden Stop Dummies and the uncertainty about the fundamentals

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

1990

:01

1990

:08

1991

:03

1991

:10

1992

:05

1992

:12

1993

:07

1994

:02

1994

:09

1995

:04

1995

:11

1996

:06

1997

:01

1997

:08

1998

:03

1998

:10

1999

:05

1999

:12

2000

:07

2001

:02

2001

:09

0

0.1

0.2

0.3

0.4

0.5

0.6

dum_SS_od s_cfw

Japan - Sudden Stop Dummies and the uncertainty about the fundamentals

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

1990

:01

1990

:08

1991

:03

1991

:10

1992

:05

1992

:12

1993

:07

1994

:02

1994

:09

1995

:04

1995

:11

1996

:06

1997

:01

1997

:08

1998

:03

1998

:10

1999

:05

1999

:12

2000

:07

2001

:02

2001

:09

0

0.2

0.4

0.6

0.8

1

1.2

1.4

dum_SS_od s_cfw

Netherlands - Sudden Stop Dummies and the uncertainty about the fundamentals

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

1995

:01

1995

:05

1995

:09

1996

:01

1996

:05

1996

:09

1997

:01

1997

:05

1997

:09

1998

:01

1998

:05

1998

:09

1999

:01

1999

:05

1999

:09

2000

:01

2000

:05

2000

:09

2001

:01

2001

:05

2001

:09

0

0.1

0.2

0.3

0.4

0.5

0.6

dum_SS dum_SS_od s_cfw

New Zealand - Sudden Stop Dummies and the uncertainty about the fundamentals

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

1994

:12

1995

:04

1995

:08

1995

:12

1996

:04

1996

:08

1996

:12

1997

:04

1997

:08

1997

:12

1998

:04

1998

:08

1998

:12

1999

:04

1999

:08

1999

:12

2000

:04

2000

:08

2000

:12

2001

:04

2001

:08

2001

:12

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

dum_SS_od s_cfw

Norway - Sudden Stop Dummies and the uncertainty about the fundamentals

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

1998

:06

1998

:08

1998

:10

1998

:12

1999

:02

1999

:04

1999

:06

1999

:08

1999

:10

1999

:12

2000

:02

2000

:04

2000

:06

2000

:08

2000

:10

2000

:12

2001

:02

2001

:04

2001

:06

2001

:08

2001

:10

2001

:12

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

dum_SS_od s_cfw

Sweden - Sudden Stop Dummies and the uncertainty about the fundamentals

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

1998

:06

1998

:08

1998

:10

1998

:12

1999

:02

1999

:04

1999

:06

1999

:08

1999

:10

1999

:12

2000

:02

2000

:04

2000

:06

2000

:08

2000

:10

2000

:12

2001

:02

2001

:04

2001

:06

2001

:08

2001

:10

2001

:12

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

dum_SS_od s_cfw

USA - Sudden Stop Dummies and the uncertainty about the fundamentals

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

1990

:01

1990

:08

1991

:03

1991

:10

1992

:05

1992

:12

1993

:07

1994

:02

1994

:09

1995

:04

1995

:11

1996

:06

1997

:01

1997

:08

1998

:03

1998

:10

1999

:05

1999

:12

2000

:07

2001

:02

2001

:09

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

dum_SS_od s_cfw

Figure 2.3: Sudden stops versus uncertainty

2.10. APPENDIX 89

Argentina - Sudden Stop Dummies and the uncertainty about the fundamentals

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

1993:03 1994:01 1994:11 1995:09 1996:07 1997:05 1998:03 1999:01 1999:11 2000:09 2001:07

0

0.5

1

1.5

2

2.5

3

dum_SS_od s_cfw

Brazil - Sudden Stop Dummies and the uncertainty about the fundamentals

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

1993

:04

1993

:09

1994

:02

1994

:07

1994

:12

1995

:05

1995

:10

1996

:03

1996

:08

1997

:01

1997

:06

1997

:11

1998

:04

1998

:09

1999

:02

1999

:07

1999

:12

2000

:05

2000

:10

2001

:03

2001

:08

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

2

dum_SS_od s_cfw

Chile - Sudden Stop Dummies and the uncertainty about the fundamentals

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

1993

:03

1993

:08

1994

:01

1994

:06

1994

:11

1995

:04

1995

:09

1996

:02

1996

:07

1996

:12

1997

:05

1997

:10

1998

:03

1998

:08

1999

:01

1999

:06

1999

:11

2000

:04

2000

:09

2001

:02

2001

:07

2001

:12

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

dum_SS_od s_cfw

Colombia - Sudden Stop Dummies and the uncertainty about the fundamentals

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

1997

:08

1997

:11

1998

:02

1998

:05

1998

:08

1998

:11

1999

:02

1999

:05

1999

:08

1999

:11

2000

:02

2000

:05

2000

:08

2000

:11

2001

:02

2001

:05

2001

:08

2001

:11

0

0.2

0.4

0.6

0.8

1

1.2

dum_SS_od s_cfw

Czech Republik - Sudden Stop Dummies and the uncertainty about the fundamentals

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

1998

:05

1998

:07

1998

:09

1998

:11

1999

:01

1999

:03

1999

:05

1999

:07

1999

:09

1999

:11

2000

:01

2000

:03

2000

:05

2000

:07

2000

:09

2000

:11

2001

:01

2001

:03

2001

:05

2001

:07

2001

:09

2001

:11

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

dum_SS_od s_cfw

Indonesia - Sudden Stop Dummies and the uncertainty about the fundamentals

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

1994

:12

1995

:04

1995

:08

1995

:12

1996

:04

1996

:08

1996

:12

1997

:04

1997

:08

1997

:12

1998

:04

1998

:08

1998

:12

1999

:04

1999

:08

1999

:12

2000

:04

2000

:08

2000

:12

2001

:04

2001

:08

2001

:12

0

0.5

1

1.5

2

2.5

3

3.5

dum_SS_od s_cfw

Korea - Sudden Stop Dummies and the uncertainty about the fundamentals

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

1994

:12

1995

:04

1995

:08

1995

:12

1996

:04

1996

:08

1996

:12

1997

:04

1997

:08

1997

:12

1998

:04

1998

:08

1998

:12

1999

:04

1999

:08

1999

:12

2000

:04

2000

:08

2000

:12

2001

:04

2001

:08

2001

:12

0

0.5

1

1.5

2

2.5

dum_SS_od s_cfw

Mexico - Sudden Stop Dummies and the uncertainty about the fundamentals

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

1993

:03

1993

:08

1994

:01

1994

:06

1994

:11

1995

:04

1995

:09

1996

:02

1996

:07

1996

:12

1997

:05

1997

:10

1998

:03

1998

:08

1999

:01

1999

:06

1999

:11

2000

:04

2000

:09

2001

:02

2001

:07

2001

:12

0

0.2

0.4

0.6

0.8

1

1.2

1.4

dum_SS_od s_cfw

Peru - Sudden Stop Dummies and the uncertainty about the fundamentals

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

1997

:08

1997

:11

1998

:02

1998

:05

1998

:08

1998

:11

1999

:02

1999

:05

1999

:08

1999

:11

2000

:02

2000

:05

2000

:08

2000

:11

2001

:02

2001

:05

2001

:08

2001

:11

0

0.2

0.4

0.6

0.8

1

1.2

1.4

dum_SS_od s_cfw

Thailand - Sudden Stop Dummies and the uncertainty about the fundamentals

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

1994

:12

1995

:04

1995

:08

1995

:12

1996

:04

1996

:08

1996

:12

1997

:04

1997

:08

1997

:12

1998

:04

1998

:08

1998

:12

1999

:04

1999

:08

1999

:12

2000

:04

2000

:08

2000

:12

2001

:04

2001

:08

2001

:12

0

0.5

1

1.5

2

2.5

3

dum_SS_od s_cfw

Turkey - Sudden Stop Dummies and the uncertainty about the fundamentals

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

1998

:05

1998

:07

1998

:09

1998

:11

1999

:01

1999

:03

1999

:05

1999

:07

1999

:09

1999

:11

2000

:01

2000

:03

2000

:05

2000

:07

2000

:09

2000

:11

2001

:01

2001

:03

2001

:05

2001

:07

2001

:09

2001

:11

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

2

dum_SS_od s_cfw

Figure 2.4: Sudden stops versus uncertainty (continued)


(1)

(2)

(3)

(1)

(2)

(3)

unce

rtai

nty

4.22

7***

4.89

4*6.

688*

0.63

2*1.

340*

1.09

2(1

.104

)(2

.644

)(3

.718

)(0

.328

)(0

.795

)(0

.795

)

mea

n ex

pect

atio

n0.

034

-0.5

65-1

.539

-0.3

13**

*-0

.697

***

-0.7

50**

*(0

.122

)(0

.425

)(1

.035

)(0

.080

)(0

.252

)(0

.261

)

lag

of R

ER

vul

n-1

.104

20.7

05-1

6.24

72.

194

64.4

80**

58.7

56**

lag

of D

LD

-14.

942*

**-4

1.94

9**

-70.

996*

*-0

.628

29.8

96*

30.1

47**

int.

RE

R v

uln

DL

D11

4.24

7***

199.

502*

*48

2.34

1**

62.7

41**

-164

.842

-155

.958

Ex.

Reg

ime

5-0

.138

0.57

12.

439

-0.1

690.

632*

0.70

0**

lag

res

over

CA

D0.

001

0.02

70.

034

-0.0

010.

023

0.01

9la

g M

2 ov

er r

eser

ves

-0.0

49*

-0.0

89-0

.575

-0.0

69*

-0.2

10-0

.224

lag

cred

it gr

owth

0.11

2-1

.818

-4.6

700.

769

4.03

02.

956

lag

FD

I/G

DP

-15.

205

-35.

652

-8.3

92-1

6.39

7*-1

9.39

5-2

4.77

4co

untr

y du

mm

ies

noye

sye

sno

yes

yes

linea

r tim

e tr

.no

noye

s**

nono

yes

quad

r. tim

e tr

.no

noye

s**

nono

yes

Con

stan

t-2

.093

**-0

.974

-5.8

74-0

.495

-7.9

47*

-8.4

27*

(0.9

11)

(2.9

17)

(9.6

58)

(0.7

64)

(4.3

13)

(4.5

07)

Obs

erva

tions

194

8484

277

137

137

Stan

dard

err

ors

in p

aren

thes

es, *

sig

nific

ant a

t 10%

; **

sign

ifica

nt a

t 5%

; ***

sig

nific

ant a

t 1%

Pool

ed P

robi

t - Y

earl

y D

ata

All

Cou

ntrie

s -

Dep

ende

nt V

aria

ble:

Sud

den

Sto

p In

dica

tor

Unc

erta

inty

Mea

sure

: Con

sens

usU

ncer

tain

ty M

easu

re: C

ombi

natio

n C

onse

nsus

WES

Table 2.4: Estimation with yearly data (all countries sample)

2.10. APPENDIX 91

(1)

(2)

(3)

(1)

(2)

(3)

unce

rtain

ty5.

890*

**9.

637*

*15

9.48

60.

814*

2.01

1*3.

497*

*(1

.82

2)(4

.028

)(7

1,8

42.5

97)

(0.4

44)

(1.0

65)

(1.7

58)

mea

n ex

p0.

311

-1.0

71-3

7.49

7-0

.244

**-0

.458

*-0

.490

(0.1

91)

(1.1

51)

(4,4

22.0

12)

(0.1

00)

(0.2

68)

(0.4

21)

lag

of R

ER v

uln

10.6

5414

.230

-899

.803

15.4

29*

67.4

34*

124.

065

lag

of D

LD-1

3.80

5*-8

0.46

6-1

,566

.751

5.28

931

.087

56.1

44in

t. R

ER v

uln

DLD

82.5

2843

8.65

99,

034.

767

12.6

53-1

61.2

71-4

55.8

48Ex

. Reg

ime

5-0

.276

1.10

950

.569

-0.0

570.

742*

0.90

4*la

g re

s ov

er C

AD

0.02

2*0.

055

-0.8

530.

008

0.01

0.04

1la

g M

2 ov

er re

serv

es0.

089

0.02

414

.904

-0.0

721.

419

3.89

9la

g cr

edit

grow

th-0

.097

3.66

615

7.02

21.

278

3.22

312

.316

lag

FDI/G

DP

-25.

805

-84.

541

-1,6

75.5

85-2

8.62

3-9

.163

-76.

614

coun

try d

umm

ies

noye

sye

sno

yes

yes

linea

r tim

e tre

ndno

noye

sno

noye

squ

adra

tic ti

me

tr.no

noye

sno

noye

sC

onst

ant

-4.8

38**

*-0

.469

-119

.757

-2.3

92-1

3.37

6-4

0.09

9*(1

.82

7)(8

.049

)(1

358

73.8

1)(1

.50

8)(8

.284

)(2

3.2

92)

Obs

erva

tions

7764

6411

597

97S

tan

dard

err

ors

in

pare

nth

eses

, * s

igni

fica

nt a

t 10

%;

** s

igni

fica

nt a

t 5%

; **

* si

gni

fica

nt a

t 1%

Pool

ed P

robi

t - Y

earl

y D

ata

Emer

ging

Mar

kets

- D

epen

dent

Var

iabl

e: S

udde

n St

op In

dica

tor

Unc

erta

inty

Mea

sure

: Con

sens

usU

ncer

tain

ty M

easu

re: C

ombi

natio

n C

onse

nsus

WES

Table 2.5: Estimation with yearly data (emerging countries sample)


(1)

(2)

(3)

(4)

(5)

(6)

lag

unce

rtai

nty

2.06

***

1.02

1***

1.06

8***

0.47

30.

503

0.55

1(0

.20

0)(0

.353

)(0

.363

) (

0.37

4)

(0.3

81)

(0.3

81)

lag

of m

ean

exp

-0.1

98**

*-0

.523

***

-0.5

44**

*-0

.638

***

-0.6

48**

*-0

.622

***

(0.0

35)

(0.0

62)

(0.0

65)

(0.0

71)

(0.0

72)

(0.0

72)

lag

of R

ER

vul

n-0

.525

3.49

43.

956

0.45

5-0

.354

0.12

1la

g of

DL

D-3

.543

***

-7.6

38**

*-8

.247

***

-4.2

96-3

.567

-3.8

95in

t. R

ER v

uln

DLD

55.4

43**

*53

.616

***

55.5

04**

*72

.544

***

78.0

18**

*73

.776

***

Ex.

Reg

ime

5-0

.153

0.13

40.

125

0.09

50.

087

0.07

1la

g re

s ov

er C

AD

0.00

6***

0.02

***

0.02

2***

0.02

3***

0.02

4***

0.02

4***

lag

M2

over

res

-0.0

03-0

.055

**-0

.054

**-0

.075

***

-0.0

83**

*-0

.071

***

lag

cred

it gr

owth

-1.8

33*

-3.2

8**

-3.5

03**

-3.9

27**

*-4

.121

***

-4.1

74**

*la

g F

DI/

GD

P1.

561

-4.0

23-5

.066

15.9

5516

.78

10.1

74co

untr

y du

mm

ies

noye

sye

sye

sye

sye

s

mon

th d

umm

ies

nono

yes

yes

yes

yes

linea

r tim

e tre

ndno

nono

yes*

**ye

sye

s

quad

ratic

tim

e tr

.no

nono

noye

s*ye

s

cubi

c tim

e tr.

nono

nono

noye

s

Con

stan

t-1

.894

-0.0

740.

134

2.49

41.

644

2.62

3

(0.2

20)*

**

(0.5

72)

(0.6

31)

(0.8

15)*

**

(0.9

35)*

(1.1

55)*

*

Obs

erva

tions

2258

1217

1217

1217

1217

1217

Sta

ndar

d er

rors

in

Pa

rant

hese

s, *

sign

ific

ant

at 1

0%

, **s

ign

ifica

nt a

t 5%

, ***

sign

ifica

nt a

t 1%

Pool

ed P

robi

t - M

onth

ly D

ata

All

Cou

ntrie

s -

Dep

ende

nt V

aria

ble:

Sud

den

Sto

p In

dica

tor

Table 2.6: Estimation with all countries sample

2.10. APPENDIX 93

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

lag

unce

rtai

nty

1.07

7***

1.36

8***

1.54

5***

1.14

0***

1.15

7***

1.18

8***

1.16

9**

0.94

1*(0

.31)

(0.3

84)

(0.4

07)

(0.4

29)

(0.4

28)

(0.4

59)

(0.4

64)

(0.4

82)

lag

of m

ean

exp

-0.2

58**

*-0

.381

***

-0.3

91**

*-0

.473

***

-0.4

58**

*-0

.497

***

-0.4

97**

*-0

.469

***

(0.0

48)

(0.0

65)

(0.0

68)

(0.0

77)

(0.0

76)

(0.0

83)

(0.0

83)

(0.0

88)

lag

of R

ER

vul

n-0

.574

0.72

50.

853

-2.2

14-2

.842

-2.2

31-2

.547

-2.7

67la

g of

DLD

-0.2

13-6

.390

**-7

.166

**-5

.511

-4.3

384.

212

3.84

58.

354

int.

RER

vul

n D

LD

44.5

32**

*58

.712

***

62.1

25**

*78

.863

***

76.9

78**

*63

.848

***

65.4

58**

*58

.846

***

Ex. R

egim

e 5

-0.1

34**

0.11

60.

10.

083

0.05

10.

111

0.11

70.

052

lag

res

over

CA

D0.

028*

**0.

016*

*0.

017*

**0.

017*

*0.

019*

**0.

032*

**0.

032*

**0.

034*

**la

g M

2 ov

er re

serv

es0.

182*

**0.

222*

*0.

247*

*0.

260*

*0.

295*

*0.

229*

0.23

5*0.

225*

lag

cred

it gr

owth

-1.0

24-2

.898

*-2

.577

-2.9

60*

-3.1

05*

-3.4

84*

-3.5

07*

-3.7

42**

lag

FD

I/G

DP

-15.

750*

**-2

.008

-3.5

312

.357

-0.7

88-2

0.18

4-1

9.75

1-2

3.25

7co

untr

y du

mm

ies

noye

sye

sye

sye

sye

sye

sye

sm

onth

dum

mie

sno

no

yes

yes*

**ye

s*ye

s***

yes

yes*

*lin

ear

time

tren

dno

nono

yes*

**ye

s*ye

s***

yes

yes*

*qu

adr.

time

trend

nono

nono

yes

yes*

**ye

sye

s**

cubi

c tim

e tr

end

nono

nono

noye

s***

yes

yes*

*fo

urth

ord

er ti

me

trend

nono

nono

nono

yes

yes*

*fif

th o

rder

tim

e tr

end

nono

nono

nono

noye

s**

Con

stan

t-1

.475

***

-1.1

63*

-1.1

460.

536

-2.4

5-2

4.87

2***

-18.

232

-266

.830

**O

bser

vatio

ns83

768

968

968

968

968

968

967

8St

anda

rd e

rror

s in

par

enth

eses

, * s

igni

fican

t at 1

0%; *

* si

gnifi

cant

at 5

%; *

** s

igni

fican

t at 1

%

Pool

ed P

robi

t - M

onth

ly D

ata

Em

ergi

ng M

arke

ts -

Dep

ende

nt V

aria

ble:

Sud

den

Sto

p In

dica

tor

Table 2.7: Estimation with monthly data (emerging countries sample)


(1)

(2)

(3)

(4)

(5)

lag

unce

rtai

nty

0.10

8***

0.06

4***

0.05

6***

0.02

5***

0.09

**(0

.037

)(0

.031

)(0

.027

)(0

.018

)(0

.068

)

lag

of m

ean

exp

-0.0

30**

*-0

.026

***

-0.0

22**

*-0

.010

***

-0.0

3***

(0.0

08)

(0.0

08)

(0.0

08)

(0.0

06)

(0.0

17)

lag

of R

ER

vul

n0.

059

-0.1

24-0

.137

-0.0

46-0

.15

lag

of D

LD

-0.5

07**

-0.3

08-0

.209

0.08

70.

68in

t. R

ER v

uln

DL

D4.

655*

**4.

411*

**3.

704*

**1.

324*

**3.

16Ex

. Reg

ime

50.

009

0.00

50.

002

0.00

20.

003

lag

res

over

CA

D0.

001*

**0.

001*

*0.

001*

**0.

001*

**0.

005*

**la

g M

2 ov

er r

eser

ves

0.01

8**

0.01

5**

0.01

4**

0.00

5*0.

013

lag

cred

it gr

owth

-0.2

30*

-0.1

66*

-0.1

49*

-0.0

72*

-0.1

46la

g F

DI/

GD

P-0

.159

0.69

1-0

.038

-0.4

19-2

.23

coun

try

dum

mie

sye

sye

sye

sye

sye

sm

onth

dum

mie

sno

yes

yes

yes

nolin

ear

time

trend

noye

sye

sye

sno

quad

ratic

tim

e tr

.no

noye

sye

sno

cubi

c tim

e tr.

nono

noye

sno

four

th o

rder

tim

e tr.

notim

e du

mm

ies

yes

Obs

erva

tions

689

689

689

689

420

Stan

dard

err

ors

in p

aren

thes

es, *

sig

nific

ant a

t 10%

; **

sign

ifica

nt a

t 5%

; ***

sig

nific

ant a

t 1%

Pool

ed P

robi

t - M

onth

ly D

ata

- mar

gina

l eff

ects

Emer

ging

Mar

kets

- D

epen

dent

Var

iabl

e: S

udde

n S

top

Indi

cato

r

Table 2.8: Marginal effects for the estimation with monthly data and emergingmarkets

2.10. APPENDIX 95

lag

expl

anat

ory

varia

bles

1. la

g2.

lag

3. la

g4.

lag

5. la

g6.

lag

(1)

(2)

(3)

(4)

(5)

(6)

lag

unce

rtai

nty

1.18

8***

1.34

5***

1.17

9***

0.82

2**

0.35

0.03

8(0

.459

)(0

.444

)(0

.42)

(0.3

98)

(0.3

9)(0

.393

)

lag

of m

ean

exp

-0.4

97**

*-0

.457

***

-0.3

95**

*-0

.320

***

-0.2

40**

*-0

.179

**(0

.083

)(0

.081

)(0

.077

)(0

.073

)(0

.071

)(0

.072

)

lag

of R

ER v

uln

-2.2

311.

692

6.29

4**

11.7

13**

*19

.231

***

26.3

15**

*la

g of

DLD

4.21

26.

564

9.00

1*12

.190

**15

.833

***

17.3

85**

*in

t. R

ER v

uln

DLD

63.8

48**

*50

.477

***

25.9

3-6

.465

-48.

207*

**-7

7.97

8***

Ex.

Reg

ime

50.

111

0.10

80.

078

0.08

10.

086

0.08

lag

res

over

CA

D0.

032*

**0.

033*

**0.

031*

**0.

032*

**0.

036*

**0.

040*

**la

g M

2 ov

er re

serv

es0.

229*

0.21

8*0.

183

0.10

9-0

.004

-0.0

63la

g cr

edit

grow

th-3

.484

*-5

.854

***

-4.8

12**

*-3

.900

***

-2.7

98**

-2.3

88*

lag

FDI/G

DP

-20.

184

-19.

906

-16.

766

-21.

202

-33.

914*

**-4

4.94

1***

coun

try d

umm

ies

yes

yes

yes

yes

yes

yes

mon

th d

umm

ies

yes

yes

yes

yes

yes

yes

linea

r tim

e tr

end

yes*

**ye

s***

yes*

**ye

s***

yes*

**ye

s***

quad

ratic

tim

e tre

ndye

s***

yes*

**ye

s***

yes*

**ye

s***

yes*

**cu

bic

time

tren

dye

s***

yes*

**ye

s***

yes*

**ye

s***

yes*

**C

onst

ant

-24.

872*

**-2

7.88

0***

-27.

670*

**-2

7.59

1***

-27.

948*

**-2

8.14

8***

Obs

erva

tions

689

680

671

662

653

644

Stan

dard

err

ors

in p

aren

thes

es, *

sig

nific

ant a

t 10%

; **

sign

ifica

nt a

t 5%

; ***

sig

nific

ant a

t 1%

Pool

ed P

robi

t Est

imat

ion

- Mon

thly

Dat

aE

mer

ging

Mar

kets

- D

epen

dent

Var

iabl

e: S

udde

n St

op In

dica

tor

Table 2.9: Different lags


(1)

(2)

(3)

(4)

(5)

(6)

pred

icte

d la

g of

unc

erta

inty

2.59

3***

3.13

0***

3.09

4***

3.50

1***

3.43

2***

3.57

6***

(0.5

9)(0

.683

)(0

.705

)(0

.677

)(0

.678

)(0

.717

)

pred

icte

d la

g of

mea

n ex

p-0

.053

-0.2

77**

-0.3

09**

*-0

.281

**-0

.281

**-0

.300

**(0

.08)

(0.1

1)(0

.114

)(0

.115

)(0

.113

)(0

.119

)

pred

icte

d la

g of

RER

vul

n16

.142

***

17.1

88**

*16

.971

***

12.0

19*

9.93

812

.477

*la

g of

DL

D2.

952

-12.

083*

*-1

2.17

6**

-12.

912*

*-1

1.82

2**

-2.4

28in

tera

ctio

n R

ER v

uln

DL

D-2

2.00

812

.921

15.7

1746

.79

52.2

6627

.906

Ex. R

egim

e 5

-0.2

02**

*-0

.023

-0.0

26-0

.011

-0.0

270.

043

lag

res

over

CA

D0.

036*

**0.

020*

**0.

021*

**0.

018*

*0.

018*

*0.

033*

**la

g M

2 ov

er r

eser

ves

0.06

90.

120.

145

0.18

7*0.

210*

0.13

2la

g cr

edit

grow

th-1

.167

-2.4

72-2

.121

-2.5

6-2

.717

-2.6

85la

g F

DI/

GD

P-2

1.97

1***

2.33

61.

418

.065

12.0

75-1

3.03

6co

untr

y du

mm

ies

noye

sye

sye

sye

sye

sm

onth

dum

mie

sno

no

yes

yes

yes

yes

linea

r tim

e tr

end

nono

no

yes

yes

yes*

**qu

adra

tic ti

me

tr.no

no

nono

yes

yes*

**cu

bic

time

tr.

nono

no

nono

yes*

**fo

urth

ord

er ti

me

tr.

nono

no

nono

notim

e du

mm

ies

nono

no

nono

noC

onst

ant

-3.5

45**

*-3

.139

***

-2.9

57**

*-2

.720

*-4

.399

**-3

1.16

1***

Obs

erva

tions

777

610

610

610

610

610

Stan

dard

err

ors

in p

aren

thes

es, *

sig

nific

ant a

t 10%

; **

sign

ifica

nt a

t 5%

; ***

sig

nific

ant a

t 1%

Inst

rum

ents

are

the

six

mon

ths

lags

of t

he u

ncer

tain

ty, o

f the

mea

n of

the

expe

ctat

ions

and

of t

he v

ulne

rabi

lity

to re

al e

The

inst

rum

ents

are

sig

nific

ant a

t lea

st a

t the

10%

leve

l in

the

first

sta

ge re

gres

sion

s

Pool

ed P

robi

t IV

Est

imat

ion

- Mon

thly

Dat

aEm

ergi

ng M

arke

ts -

Dep

ende

nt V

aria

ble:

Sud

den

Stop

Ind

icat

or

Table 2.10: IV estimation

2.10. APPENDIX 97

Met

hod

logi

tco

nditio

nal lo

git,

feC

ham

ber

lain

's

pan

el p

robi

t(1

)(2

)(3

)(4

)(5

)

lag

unce

rtai

nty

1.1

57*

**1.

188

***

2.07

2**

1.9

86**

1.1

88**

* (

0.42

8)

(0.

459)

(0.8

46)

(0.8

33)

(0.4

59)

lag

of m

ean

exp

-0.4

58*

**-0

.497

***

-0.9

20**

*-0

.88

8***

-0.4

97*

**(0

.076

)(0

.083

)(0

.157

)(0

.153

)(0

.083

)

lag

of R

ER

vul

n-2

.842

-2.2

3-4

.11

1-3

.86

3-2

.23

1

lag

of D

LD-4

.338

4.2

127.

987

7.80

84.

212

int.

RE

R v

uln

DLD

76

.98*

**63

.85

***

118

.33

***

113

.38

***

63.

85*

**

Ex.

Reg

ime

50.

051

0.1

110

.15

50

.14

60.

111

lag

res

over

CA

D0

.01

9***

0.12

7**

*0.

056

***

0.0

54**

*0

.032*

**

lag

M2

ove

r re

serv

es0

.29

5**

0.2

29*

0.3

99*

0.3

73

0.2

29*

lag

cred

it gr

ow

th-3

.105

*-3

.48

4*-7

.421

**-7

.466

**-3

.484*

lag

FD

I/GD

P-0

.788

-20.

184

-36

.482

-35

.42

0-2

0.1

84

coun

try

dum

mie

sye

sye

sye

sno

no

mon

th d

umm

ies

yes

yes

yes

yes

yes

linea

r tim

e tr

end

yes

yes

yes*

**ye

s***

yes*

**

qua

dra

tic ti

me

tr.

yes*

yes

yes*

**ye

s***

yes*

**

cubi

c tim

e tr

.no

yes

yes*

**ye

s***

yes*

**

Con

stan

t-2

.45

-24

.87

2***

-46.

077

***

no

-28

.798

***

Ob

serv

atio

ns68

96

896

8968

983

7St

anda

rd e

rror

s in

Par

anth

eses

, *si

gnifi

cant

at 1

0%, *

*sig

nific

ant a

t 5%

, ***

sign

ifica

nt a

t 1%

pool

ed p

rob

it

Em

ergi

ng M

arke

ts -

Dep

end

ent V

aria

ble:

Sud

den

Sto

p

Ind

icat

or

Table 2.11: Alternative estimation methods


Focal crises - Headline crises - (large IMF packages, defaults, currency crises) measures

Private Net Flows on Debt 5/

Country year What defined crisesIMF-supported Programs/Aid packages

((GDPt - GDPt-1) /GDPt-1)*100

((GDPt+1 - GDPt-1) /GDPt-1)*100,

Global Development

Finance 2/ WEO 3/ BOP YB 4/ WEO 3/ BOP YB 4/

the 70s

Peru 1978sovereign default, Currency crisis (FR), no banking crises

1978: IMF stabilization program+ multilateral rescheduling with official and private creditors 0.08 5.68 -6.89 -4.75 -5.93 -4.32 -4.77

Turkey 1978sovereign default, no currency crises, fall in Central bank reserves, 1982-85 Systemic banking crisis 2.83 1.93 -5.89 -2.65 -1.61 -2.88 -2.01

United Kingdom 1974-76Currency Crisis in 1976 (ERW), Borderline and smaller banking crisis -1.70 -2.38 5.08 2.24 5.08 2.25

Zaire 1978

Sovereign default since 1976, Enormous amounts of external debt lead to Paris Club reschedulings in 1979 as well as 1981 and with a syndicate of commercial banks in 1980, Currency Crisis in 1979 (MR and BP), 1980s Systemic banking crisis -5.30 -5.02 -0.21 -6.86 -6.69

crises countries 80s

Argentina 1982-88 1982

sovereign default, Currency Crises in 1981 (MR1, FR, GKR) and 82 ( FR, BP, GKR), 1980-82 Systemic banking crisis, 1989-90 Systemic banking crisis, hyperinflation -3.15 0.47 0.24 0.18 -0.11 0.16 -0.23

Bolivia 1980

sovereign default in 1980, hyperinflation, Spring 1984 suspension of interest payments to commercial banks, Currency Crises in 1980 (MR2), 1982 (MR, FR, BP, GKR), 83 (FR, BP, GKR), 84 (FR) and 85 (FR, BP, GKR),1986-88 Systemic banking crisis 0.61 1.54 -10.13 -12.29 -2.40 -19.60 -14.72

Brazil 1982sovereign default 1983, Currency Crisis in 1982 (BP) and 83 (FR, BP, GKR), no banking crisis

Brady Plan: Brazil Parallel Financing agreement, terms announced Sep 1988 -4.36 -8.63 0.24 -0.69 -0.41 -1.12 -0.38

Bulgaria 1990 1989

No sovereign default but during second half of 80s build up of large external debt in order to finance enlarging current account deficit. no data on currency crises available, but exhaustion of foreign reserves. 1995-97 Systemic banking crisis

Brady Plan: Bulgaria Brady, terms announced Mar 1994 -0.50 -9.55 0.68 -0.30 -3.88 0.49 -2.01

Chile (Cline p. 287, http://www.pbs.org/wgbh/commandingheights/lo/countries/)1982

Sovereign default in 1983, Currency Crises in 1982 (MR, FR, GKR) and 83 (FR), 1981-86 Systemic banking crisis -13.42 -16.44 -5.50 -9.57 -10.03 -9.95 -10.49

China 1990Currency Crises 1990 (MR), 1991 Systemic banking crisis 3.80 13.35 1.14 -1.74 -0.21 -2.46 -0.95

Colombia (Cline p.280) 1983No Sovereign default, Currency Crises in 1983 (GKR) 1985 (BP and GKR), 1982-87 Systemic banking crisis 1.57 4.98 -1.88 -3.21 -2.56 -3.72 -3.33

Costa Rica 1981

Sovereign default, Severe balance of payment crisis, Currency Crises in 1981 (MR and FR), no banking crisis

Brady plan: Costa Rica Brady terms announced May 1990 0.80 -6.25 -8.41 -7.07 -10.07 -9.32 -7.19

Cote d'Ivoire 1984Sovereign default, no Currency Crisis, Systemic banking crisis from 1988-91 Brady plan concluded in 1997 -2.00 1.55 0.38 -25.75 -13.32 -20.19 -12.80

Ecuador 1982

Sovereign default, Currency Crises in 1983 (MR2), 84 (MR1) and 86 ( MR, FR),1980- 83 Systemic banking crisis

Brady plan:Ecuador Brady, terms announced May 1994 1.20 -1.63 -7.93 0.93 7.40 3.46 2.30

Korea (Sachs, p. 121) 1979-821980

No Sovereign default, but in 1981world's fourth largest debtor country. Currency Crisis in 1980 (MR, BP and GKR), Doubling of inflation from 14.4 % in 1978 to 28.7 % in 1980. -2.09 4.24 -0.11 0.77 1.99 6.74 1.83

Jordan 1989

Sovereign default on loans to commercial banks, Currency Crisis (MR1, FR, BP), Non systemic banking crisis

Brady Plan: Jordan Brady, terms announced in July 1993 -13.45 -7.29 24.74 -41.37 -3.59 -34.87 -5.61

OutputNet Private Capital

Flows

Net Private Capital Flows plus Net Errors

and Omissions

Table 2.12: Headline crises from 1970 - 2000

2.10. APPENDIX 99

Focal crises - Headline crises - (large IMF packages, defaults, currency crises) measures

Private Net Flows on Debt 5/

Country year What defined crisesIMF-supported Programs/Aid packages

((GDPt - GDPt-1) /GDPt-1)*100

((GDPt+1 - GDPt-1) /GDPt-1)*100,

Global Development

Finance 2/ WEO 3/ BOP YB 4/ WEO 3/ BOP YB 4/

the 90s till present

Argentina 1995

Contagion from Mexican crisis, background currency board without deposit incsurance scheme and without lender of last resort: withdrawal of bank deposits, significant loss of central bank's gross reserves, liqidity crunch, surge in interest rates, output contraction, Systemic banking crisis X / EFF 3.85 7.79 5.36 -1.16 -1.88 -1.91 -2.38

Argentina 2001-2sovereign default, no data on currency crisis, 2001-present Systemic banking crisis 2.10 5.27 -7.19 -4.28 -5.02 -5.06

Brazil 1998

No sovereign default, Currency crises in 1999 (MR1), substantial curret account deficit, surge of interest rates, outflow of capital, Output contraction, 1994-9 Systemic banking crisis

X/ SBA/ SRF, new arrangement, 12/2/98, X / SBA/SRF, new arrangement, 9/14/01, 0.13 0.92 0.07 0.41 0.82 0.31 0.85

Ecuador 1999

El Nino crisis, default on external and internal debt, Currency Crises in 1999 (MR1), 1998-present Systemic banking crisis -6.30 -3.67 -4.26 -16.63 -13.47 -14.09 -15.36

ERM 1992/1993

Currency Crises: Denmark 93 (GKR), Finland 92 (GKR, ERW), Ireland 92 (ERW), Italy 92 (ERW), Portugal 93 (MR1), Spain 92 (GKR) and 93 (GKR), Sweden 92 (ERW), UK 92 (ERW) 0.00 0.00

Finland 1991-94No sovereign default, Currency Crisis in 1991 (GKR) and 92 (GKR, ERW), Systemic banking crisis -6.26 -9.37 -3.20 -5.88 -1.55 -4.19

Indonesia 1997-98

no sovereign default, consequence of unresolved capital account crisis, important short-term private sector external debt, depreciation, hyperinflation, runs on deposits, collaps of corporate balance sheets, sharp economic contraction, Currency Crisis in 1997 (MR2, BP, GKR), 1997-present Systemic banking crisis

X/ SBA, new arrangement, 11/5/97 4.54 -9.18 -2.92 -0.90 -3.71 0.95 -5.46

Korea 1997-8

No sovereign default, but high level of short term private foreign debt, Curreny Crisis in 1997 (MR2, FR, BP, GKR) and 98 (MR1)

X/ SBA/SRF, new arrangement, 12/4/97 5.01 -2.01 -6.32 -8.52 -4.54 -9.70 -5.72

Mexico 1994-5

Tequila crisis. No sovereign default, Currency Crisis in 1994 (BP, GKR) and 95 (MR), 1994-97 Systemic banking crisis

X / SBA, new arrangement, 2/1/95 -6.17 -1.33 0.10 -4.09 -4.32 -2.67 -2.90

Malaysia 1997-8

No sovereign default, interest rate surge, real GDP contraction, Currency Crisis in 1997 (FR, BP), 1998 (MR1), 1997-present Systemic banking crisis 7.32 -0.57 1.39 -7.82 -7.17 -7.55 -4.83

Norway 1987-93 1989No sovereign default, currency crisis in 1986 (ERW), Systemic banking crisis 0.90 2.92 -2.69 -2.94 -2.85 -3.10

Pakistan 1999-2000Sovereign default, Eurobond exchange, no Currency Crisis in 1999, 2000 n.a., no data on banking crisis 3.96 7.57 -0.72 -1.06 0.97 -0.49 0.57

Phillipines 1997No sovereign default, Currency Crisis in 1997 (FR, GKR), 1998-present Systemic banking crisis X/ EFF 5.19 4.58 2.55 -4.83 -4.82 -7.60 -7.54

Russia 1998

Sovereign default 1998-99, interest rate surge, Currency Crises in 1998 (GS), 1998-9 Systemic banking crisis

EFF/SFR/CCFF, Augmentation and Extension, 7/20/98 -4.90 0.24 2.90 0.26 1.46 -0.51 0.47

Sweden 1991No sovereign default, Currency Crisis 1992 (GKR), Systemic banking crisis -1.11 -2.83 -4.02 -9.30 -4.02 -4.67

Thailand 1997-8

No sovereign default, but roll over of short term debt stopped, Currency Crisis in 1997( MR2, FR, BP, GKR), 1997-present Systemic banking crisis

X/ SBA, new arrangement, 8/20/97 -1.37 -11.74 -7.11 -18.42 -12.75 -18.54 -13.05

Turkey 1994

No sovereign default, interest rate surge, Currency Crisis in 1994 (BP, GKR), Non systemic banking crisis X/ SBA -4.97 1.57 -9.47 -4.45 -6.19 -4.45 -4.23

Turkey 2000No sovereign default, no data on currency crisis available, 2000-present Systemic banking crisis

X / SBA/SRF, augmentation, 5/15/01, SBA, new arrangement, 2/4/02 7.36 -0.69 1.18 -1.33 3.75 -1.33 1.53

OutputNet Private Capital

Flows

Net Private Capital Flows plus Net Errors

and Omissions

Table 2.13: Headline crises from 1970 - 2000 (continued)

Chapter 3

The Uncertainty About theFundamentals and the Spread ofStock Market Crises

3.1 Introduction

Financial crises in emerging markets in recent years have been especially cen-

tered around the Mexican (December 1994), the Thai (July 1997), and the

Russian (August 1998) crises. Financial markets witnessed a similar accu-

mulation of crises in developed countries in the context of the crisis of the

European Exchange Rate Mechanism (September 1992).1 These periods of

crises concentration suggest contagion effects across countries.

Because of the high costs of these financial crises in emerging markets,

researchers and practitioners have been exploring these cases. Specifically un-

der investigation are the mechanisms through which crises spread, the factors

that render countries vulnerable to contagion, and, most importantly, which

policies might help prevent contagion.

This chapter addresses these questions by analyzing one particular mecha-

nism of the spread of crises: the contagion of crisis through uncertainty about

the fundamentals. This chapter focuses on financial crises characterized by a

severe plunge in stock market returns.

1See, for example, Broner et al. (2006), Caramazza et al. (2004), or Kaminsky et al.(2000) for the dates of the crises.

101

102 CHAPTER 3. UNCERTAINTY AND SPREAD OF CRISES

It can be observed that, after a number of crises, the disagreement about

the fundamentals in other markets – especially those markets that are later on

themselves hit by a crisis – increases. Figure 3.1 illustrates this observation in

the case of the Thai crisis in 1997.2 As illustrated in the graph, the uncertainty

not only increases in Thailand after the crisis, but also in neighboring countries.

Korea, for example, is characterized in the data by a build-up of uncertainty

after the Thai crisis. Korea is then hit by a currency crisis in November 1997.

In addition, Figure 3.1 shows that the crisis in Thailand does not have an

effect on the degree of uncertainty in Taiwan and the UK, neither were these

countries economically strongly affected by the crisis.

However, in the case of other financial crises, careful scrutiny reveals that

uncertainty about the fundamentals decreases in other markets after the crisis

in the initial market. For example, this is the case in the period around the

Argentinean crisis in 2002, which is illustrated in Figure 3.2.

The recent literature distinguishes between surprise crises as, for example,

the Thai crisis in 1997 and anticipated crises as, for example, the Argentinean

crisis in 2001/2002. This literature argues that the international repercussions

of the anticipated crises in Brazil (January 1999), Turkey (February 2001),

and Argentina (December 2001) were much less important than those after

the crises in Mexico (December 1994), Thailand (August 1997), and Russia

(August 1998).3

This chapter picks up this distinction and shows that surprise crises in-

crease uncertainty about fundamentals in other countries, thereby resulting in

a higher probability of crises there. In contrast, the occurrence of anticipated

crises decreases disagreement about the state of the fundamentals in other

countries, thereby lowering the probability of a crisis there.

This chapter contributes to the literature in two ways: First, uncertainty

about the fundamentals is theoretically illustrated as a factor transmitting

2The left Y-axis displays the crisis variable. The two bars in the figure show the two mostpronounced crisis events in the Thai crisis: First, the severe devaluation of the Bath in thebeginning of July 1997 and second, the substantial drop in stock market returns one monthlater. The dates are chosen in accordance with Kaminsky et al. (2000) and Goldstein (1998).The right Y-axis displays the uncertainty about the fundamentals in the tracked economies.Uncertainty is measured by the standard deviation of growth forecasts for the current andfollowing year, by financial analysts within the tracked countries. For more details on thismeasure, please refer to Appendix 3.6.

3See Kaminsky et al. (2000) or Didier et al. (2006)


0

1

1997:01

1997:02

1997:03

1997:04

1997:05

1997:06

1997:07

1997:08

1997:09

1997:10

1997:11

1997:12

1998:01

1998:02in

cid

ence

s o

f cr

isis

0

0.5

1

1.5

2

2.5

3

3.5

un

cert

ain

ty

crisis in Thai Thailand Indonesia

Korea Malaysia Hong Kong

Singapore Taiwan United Kingdom

Devaluation of the Thai bath

Biggest drop in stock market returns

Figure 3.1: Uncertainty in the surroundings of the Thai crisis

crises across markets. Second, predictions of the theoretical model are vali-

dated empirically. The role of uncertainty about the fundamentals has been

neglected in the existing literature on contagion. So far, common investors

have been detected as the main reason of the spread of financial crises between

economies. While early research focused on trade linkages4 and on macroeco-

nomic similarities between economies5, more recent analyses converge to the

view that common creditors are at the core of contagion. This view is sup-

ported by a large number of empirical analyses.6

Based on the insight into the role of common investors, the theoretical liter-

ature suggests different propagation mechanisms. Research thus far examines

4See, for example, Gerlach and Smets (1996).5See, for example, Goldstein (1998).6See, for example, Van Rijckeghem and Weder (2001) and Caramazza et al. (2004).


0

1

2001:06

2001:07

2001:08

2001:09

2001:10

2001:11

2001:12

2002:01

2002:02

2002:03

2002:04

2002:05

2002:06

inci

den

ces

of

cris

is

0

0.5

1

1.5

2

2.5

3

un

cert

ain

ty

crisis in Arg Argentina Brazil Mexico

Chile Columbia United Kingdom Peru

announcement of default

Figure 3.2: Uncertainty in the surroundings of the Argentinean crisis

herding due to fixed information cost7, differently informed investors8, changes

in investors’ risk aversion9, and wealth effects10 as possible propagation chan-

nels for crises.

Following Goldstein and Pauzner (2004), I model the financial crisis in

country B as a coordination game between private investors. The reason for

using a coordination game is that the setup of a coordination game is well

suited to analyze the effect of uncertainty about the fundamentals. The present

model differs from the Goldstein and Pauzner (2004) setup in two crucial

ways. The first difference concerns modeling the initial-crisis country and

the potentially-affected subsequent country. While Goldstein and Pauzner

7See Calvo and Mendoza (2000).8See Kodres and Pritsker (2002).9See Broner et al. (2006).

10See Goldstein and Pauzner (2004).


(2004) explicitly model the sequence of two bank-run crises of the Diamond

and Dybvig (1983) type, I focus on the second economy exclusively. I model

the occurrence of a crisis in the second country, assuming that either a surprise

crisis takes place in the first country or an anticipated crisis.

The second difference concerns the mechanism through which the crisis

spreads. In Goldstein and Pauzner (2004), the crisis spreads due to a wealth

effect. In my setup, the change in uncertainty about the fundamentals trans-

mits the crisis. I assume that uncertainty about the fundamentals increases in

the second country if a surprise crisis hits the first country. Further, I assume

that uncertainty decreases if an anticipated crisis occurs in the first country.

The illustrative model in this chapter is then used to show that an increase in

uncertainty increases the probability of a crisis in the second country while a

decrease in uncertainty makes a crisis less likely there.

This study offers two justifications of the assumption that a surprise crisis

in an initial-crisis country increases the uncertainty in another country: The

first justification is the empirical evidence presented in Figures 3.1 and 3.2.

The second justification is the following line of arguments: If a crisis hits a

country by surprise, i.e., without investors expecting the event, investors learn

that they did not put sufficient effort into information processing given existing

data-processing technology. If they want to predict crises in other countries,

they have to increase their investment in information processing. However,

a number of the investors realizes losses in the first country and, hence, are

less inclined to invest in the second economy.11 Given the assumption that the

payoff of one agent positively depends on the fraction of other agents investing,

i.e., that strategic complementarity prevails between investments, this leads to

all agents optimally choosing to spend less on their information processing after

the crisis in the first country.12 As a result, all agents receive more dispersed

signals about the true value of the fundamentals.

This mechanism about how the degree of uncertainty depends on a cri-

sis in a first country works in the opposite direction if an anticipated crisis

materializes. In this case, investors’ trust in their information processing is

strengthened and they are willing to spend a higher amount on gathering infor-

mation, despite the crisis in the first market. This higher effort in information

processing, in turn, leads to more precise signals.

11This is an outcome of the model by Goldstein and Pauzner (2004).12The assumption of strategic complementarity is common in the global game literature.


The model in this chapter illustrates the presence of contagion in a coordi-

nation game: In country B, infinitely many investors (agents) have one unit of

endowment available for investment there. If they choose not to invest, they

receive a certain return of zero. In case that they invest, the return depends

positively on the fraction of other agents who invest. In addition, the return

decreases with an increasing level of the fundamentals. A high level of funda-

mentals indicates high costs of investing (this could be due to high political

instability or high transaction costs). The fundamentals of the economy are

uniformly distributed over a finite support. However, investors cannot observe

the true realization of the fundamentals but receive a private signal that is

symmetrically and uniformly distributed around the realization of the true

fundamentals. This means that investors base their investment decisions on

the expected return, given their private evaluation of the fundamentals.

This information structure yields a threshold equilibrium in terms of the

fundamentals in B and the outcomes in A. Below the threshold, the investors

coordinate on investing; above, no one invests. Comparative static analysis

shows that the threshold is a decreasing function of the dispersion of the private

signals. The dependence of the uncertainty in B on the crisis in A together

with the result of the comparative static analysis of the threshold in B are

sufficient to illustrate the existence of contagion: A surprise crisis in country

A increases the dispersion of the private signals, i.e., the support of the private

signals around the true value of the fundamentals, in B and hence, decreases

the threshold there. The decrease in the threshold means that coordination on

the bad equilibrium becomes more likely, i.e., a crisis becomes more probable.

In the case of an anticipated crisis in country A, the opposite is true.

To validate empirically the uncertainty channel of contagion, I construct

a rich data set for 38 countries with monthly time series (December 1993 to

September 2005). This country sample and the associated time frame enables

the inclusion of the following six pronounced crisis periods into the analysis:

Mexico (1994), Thailand (1997), Russia (1998), Brazil (1999), Turkey (2001)

and Argentina (2001). The two main variables in the data set are a stock

market crisis dummy and an uncertainty measure. A stock market crisis is de-

tected by significant negative variation in stock market returns. The monthly

stock market returns that serve as a basis for the crisis dummy are computed

from the IFC (International Finance Corporation) investable US dollar total


return index.13 When necessary, I complete the returns with data from MSCI

(Morgan Stanley Capital International) or national sources. As in the pre-

vious chapter, the measure of uncertainty is the standard deviation of GDP

growth forecasts between country experts. Additionally, I employ a large set

of domestic control variables and alternative channels of contagion.

I proceed in two distinct steps. Firstly, I use fixed-effects panel estimations

to establish the link from the initial crisis in country A to the uncertainty

in other countries B. I control for country and time effects, running vari-

ous robustness checks. Secondly, I quantify the effect of uncertainty in those

economies on the probability of a crisis there. For this second step, I employ

pooled probit estimation, controlling for country and time effects. Again, I

control for potential domestic drivers of crises. Finally, as a check for alterna-

tive channels of contagion, I control for contagion through common creditors,

trade links, the size effect of the initial stock market, and for overexposed

common fund investors.

The empirical analysis in this chapter expands the existing empirical liter-

ature on the spread of crises in several respects. First, the effect of uncertainty

in the context of the spread of crises has been neglected so far. Second, as the

panel data spans a larger time horizon, I can consider a larger number of crises

periods.14 Third, I control for a large number of alternative contagion channels,

adapting them to the particular kind of crises analyzed – namely, substantial

stock market drops. Fourth, including control for time effects results in very

strict tests for the transmission channels of crises. The time-effects control

takes care of all effects present at a particular point in time. In case of all

emerging markets, the time-effects control for increases in the interest rates in

the financial centers. Not all of the alternative contagion channels controlled

for remain significant when controlling for time effects.

The analysis yields two main empirical findings. The first finding is that

uncertainty about the fundamentals is a propagation mechanism of contagion,

if the first country is hit by a surprise crisis. The first step of the analysis

finds that the Mexican, Thai, and Russian crises increase the uncertainty in

potentially-affected countries. The effect is stronger within the region where

13The investable indices take into consideration restrictions on foreign investment. There-fore, this measure represents the part of the national stock markets accessible to foreigninvestors, which is relevant in the context of contagion.

14For example, Van Rijckeghem and Weder (2001) only consider the Mexican, Thai, andRussian crises, while Broner et al. (2006) analyze the Thai, Russian, and Brazilian crises.


the crises occur; the effect appears more pronounced in countries nearer to the

initial crises country. The second step of the analysis finds that the effect of

uncertainty on crisis probability in countries B is positive, significant, and, as

shown by marginal effects, not negligible in size.

The second finding is that in the case of an anticipated crisis, uncertainty

about the fundamentals in the second economy is decreased, which, in turn,

decreases the probability of a crisis there. The first step of the analysis yields

the following result: The Brazilian, Turkish, and Argentinean crises decrease

the uncertainty in the potentially-affected countries. The effect is stronger

within the region where the crises occur and in countries closer to the initial-

crisis country. The second step of the analysis confirms that the effect of

uncertainty in the potentially-affected countries on the probability of a crisis

there is positive, significant, and not negligible.

These findings have several implications. The first, obvious implication is

that a close monitoring of the fundamentals in the resident country and also

of other countries is crucial. Particularly other countries in the first-country

region and geographically close ones should be focused on. Surprise crises seem

to be especially bad because they set off mechanisms that further worsen the

situation. This chapter illustrates such a mechanism through the uncertainty

about the fundamentals. The second implication is that, once a surprise crisis

has hit a first country, policy makers in potentially-affected countries should

move toward policies that diminish the potential increase in uncertainty. One

venue could be to develop mechanisms for such situations through which gov-

ernments could credibly disseminate very precise information about the state

of their economy so that the private signals get as precise as possible. One

could even start to think about subsidies for information-gathering technology.

The chapter is organized as follows. In section 3.2, I describe the model.

In section 3.3, I present the empirical analysis. Section 3.4 explains policy

implications while section 3.5 contains the conclusion.

3.2 The Model

This section presents a simple coordination game to illustrate the occurrence

of contagion between two markets that are uncorrelated in terms of their fun-

damentals. The focus of the model is on the potentially-affected country. In

3.2. THE MODEL 109

an investment game, I illustrate that a crisis in country B becomes more prob-

able after a surprise crisis in country A and becomes less probable after an

anticipated crisis materializes in a first country. The transmission functions

through the uncertainty about the fundamentals. For the theoretical illustra-

tion of this channel, three ingredients are necessary. A first ingredient is that

the dispersion of private signals in country B increases due to a surprise crisis

in country A and, conversely, that dispersion of private signals decreases due

to an anticipated crisis in country A. In this study this effect of a crisis on

uncertainty about the fundamentals is introduced as an assumption.15

The second ingredient is a unique threshold equilibrium in terms of the

fundamentals of the economy, so that it is possible to attribute to each level

of the fundamentals the realization of the investment or the non-investment

equilibrium. Once this unique threshold equilibrium is determined, the third

ingredient is the comparative static analysis of the threshold equilibrium with

respect to the uncertainty about the fundamentals. If the threshold shifts with

changes of the uncertainty, contagion is present.

The assumption that a surprise crisis in an initial-crisis country increases

uncertainty about the fundamentals in another country can be justified by the

empirical evidence presented in the introduction to this chapter. Addition-

ally, it could be argued that investors learn after a surprise crisis that they

did not put sufficient effort into information processing, given existing data-

processing technology. If investors want to predict crises in other countries,

they must increase investment in information processing. However, a number

of the investors realize losses in the first country and, hence, are less inclined

to invest in the second economy.16 Given the assumption that the payoff of

one investor depends positively on the fraction of other agents investing, i.e.,

that strategic complementarity prevails between investments, this leads to all

investors choosing optimally to spend less on information processing after the

crisis in the first country.17 As a result, all investors receive more dispersed

signals about the true value of the fundamentals. The mechanism works in

the opposite direction if an anticipated crisis materializes in the initial-crisis

country. In this case, investor trust in information processing is strengthened.

Therefore, investors are willing to spend a higher amount on gathering infor-

15It is an interesting topic for future research to explicitly model this effect.16This is an outcome of the model by Goldstein and Pauzner (2004).17The assumption of strategic complementarity is common in the global game literature.


mation despite the crisis in the first market. This in turn leads to more precise

signals.

The notion of a surprise crisis and an anticipated crisis in the first crisis

country are absent in the setting of Goldstein and Pauzner (2004). However,

the idea of the distinction between surprise crises and anticipated crises is

consistent with the setup of a global game. Think of a surprise crisis in the

following way: If the prior expectation about the value of the true fundamentals

is lower than the threshold equilibrium, investors, on average, expect that no

crisis will happen. If the fundamentals are then realized in the range above

the threshold, this realization can be interpreted as a surprise crisis. On the

other hand, if the prior expectation about the value of the true fundamentals

is higher than the threshold equilibrium, investors expect the bad equilibrium

to be realized. If is the bad equilibrium is then realized, this can be interpreted

as an anticipated crisis.

3.2.1 Model Setup

Here, I describe the game in country B taking as a given the outcomes in the

initial crises country A.18

As with the fundamentals in the previous chapter, in this chapter the in-

vestment environments are assumed to be uniformly distributed over the finite

interval θ ∼ [θ, θ]. A high value of the fundamentals θ signifies an adverse

environment for investment with high investment obstacles.

There is a continuum of [0, 1] identical investors . Each investor decides

whether to invest 1 unit or not. If an investor does not invest, he receives a

certain return of 0. If he decides to invest, he receives an uncertain return

of P (θ, π−i), which depends negatively on the level of fundamentals θ and

positively on the fraction of other investors that invest in B, π−i. A strategy

is defined as πi : [θ, θ] → [0, 1], which means that investor i invests in state

θ with probability πi(θ). Due to the mass of agents being 1, the fraction of

agents who invest at a particular state of fundamentals can be expressed as∫ 1

0πj(θ)dj = π−i(θ) for j 6= i. The positive dependence on the fraction of other

agents investing, i.e., strategic complementarity between the agents, can be

18The investment game in country B is a straight application of the theory of globalgames by Carlsson and van Damme (1993). Similar investment games have been used inthe literature, for example, by Heinemann (2005).

3.2. THE MODEL 111

explained by increasing returns on aggregate investment. These assumptions

are reflected by the following payoff function of investor i:

P = Rπ−i(θ)− 1

2θ2 (3.1)

in which Rπ−i(θ) stands for the simplest form of a return that is positively

dependent on the fraction of other agents investing. Further, the last term

can be interpreted as a cost of investing that increases exponentially with the

worsening of the investment environment.

An investor decides whether to invest or not to invest in a country after

receiving information about the fundamentals of the country. Assuming that

the fundamentals are not common knowledge, each investor privately inter-

prets publicly available information. The investors thus act upon their private

signals. The private signals are uniformly distributed in an η surrounding of

the true fundamentals θ: θi ∼ U [θ− η, θ + η]. Now, the variance of the signals

depends on the outcomes in country A. In the case of a surprise crisis in coun-

try A, the private signals are uniformly distributed in an η + c surrounding of

θ, with c being a small positive number. In case of an anticipated crisis in A,

the private signals are uniformly distributed in an η−d surrounding of θ, with

d being a small positive number.

An investor is more likely to invest if 1) the obstacles to invest are lower and

2) if a large number of other investors invest in the same country. However, in

line with global games literature, I assume that there are small ranges at the

extremes of the support of the fundamentals where investors have dominant

strategies. If the fundamentals are very good, i.e., if the investment obstacles

are very low, it is optimal for an investor to invest irrespective of the actions of

all the other investors. On the other extreme, if the state of the fundamentals

is very adverse to investment, then it is the optimal strategy of an investor not

to invest, irrespective of the actions of the other investors.

Formally, this assumption means that the support of the fundamentals has

to exceed the border of the dominance region by at least 2η: θ + 2η < θ <

θ < θ − 2η, in which θ stands for the border of the lower dominance range

and θ stands for the border of the upper dominance range. This condition

ensures that an investor is indifferent between investing and not investing at

the borders of the dominance ranges. At the border of the dominance region

at the high end of the support, the investor is indifferent, even if the fraction of

other agents investing equals 1, P (θ, 1) = 0. At the border at the dominance


region at the low end of the support, the investor is indifferent even if no

one else invests, P (θ, 0) = 0. When an investor receives a signal θi < θ − η,

he knows that his payoff P i > 0, no matter what all the other investors are

doing. Therefore, he will invest. Analogously, if he receives a private signal

θi > θ + η, he knows that P i < 0, no matter what all the other investors are

doing. Therefore, he will not invest. In contrast, between the borders of the

dominance regions, the payoff of an investor depends on the actions of other

investors. This results in the same tripartite partition of the fundamentals as

in the model of chapter 2. Under common knowledge, multiple equilibria exist

in this intermediate range of fundamentals. Because the common knowledge

game is extensively described in chapter 2, I proceed here immediately to the

investment game with private information.

3.2.2 Solving the Model

Firstly, I will show that the game with private information is characterized by

a unique Bayesian Nash equilibrium in country B.

Proving the existence and the uniqueness of the equilibrium requires several

steps. In the first step, a simple switching strategy is assumed to be followed

by all investors. In the second step, the monotonicity of the expected payoff

difference in the private signal has to be proved. Based on this, dominated

strategies can then be iteratively eliminated in the third step, beginning at

borders of the dominance regions. Finally, it has to be shown that there is

only one unique value of the level of debt, for which the payoff difference,

given the private signal, equals zero. This level of debt is the threshold value,

below which all agents invest and above which no one invests.

In the private information game, a strategy is a function of the private signal

received instead of the true value of the fundamentals: πi(θi) : [θ, ˆtheta] →[0, 1].19 The payoff function of an investor now depends on his private signal

on the state of the fundamentals and is therefore given by

P (θi) = E[Rπ−i(θj)− 1

2θ2|θi] (3.2)

19Note that the private signal is drawn from the same support as the true value of thefundamentals. No private signal will be realized at a level of debt that is, in reality, nonex-istent. As noted earlier, the support of the true value of the fundamentals must exceed theborders of the dominance regions sufficiently, i.e., by 2η, so that there exist private signalsthat are consistent with those dominant strategies.

3.2. THE MODEL 113

Analogously, the fraction of other agents investing πi(θi) is a function of

the private signals θj they receive.

In the first step towards the unique equilibrium, it is assumed that all

investors follow a simple switching strategy. A switching strategy IT means

that an investor invests with probability one if, and only if, the signal it receives

is below a threshold T and abstains from investing with probability one if the

signal is above the threshold20

IT =

1 if θi < T0 if θi ≥ T

(3.3)

The simple switching strategy permits rewriting the payoff function, re-

placing the fraction of other investors investing with the probability that one

other investor receives a signal that is smaller than the threshold signal

π−i(IT ) =

∫ 1

0

IT (θj)dj = prob(θj ≤ T ) (3.4)

P (θi, IT ) = R · 1 · prob(θj ≤ T ) + R · 0 · prob(θj > T )− E(1

2θ2|θi) (3.5)

Recall that at the borders of the dominance regions, the investors are in-

different between investing and not investing.21 If the payoff function is mono-

tonically decreasing in the private signal, clearly, these borders are the lowest

and the highest possible threshold signals for the switching strategies. In the

dominance region at the low end of the support of the fundamentals, the invest-

ment obstacles are so low that the payoff of an investor is positive if investing,

irrespective of the actions of all other investors. At the border itself an investor

is, then, indifferent. In the dominance region at the high end of the support,

the investment obstacles are so high that the payoff of an investor is positive

if investing, irrespective of the actions of all other agents. In the case of a

monotone payoff function, the borders of the dominance regions are, therefore,

the starting points of the iterative elimination of dominated strategies.

Accordingly, in a second step towards the unique equilibrium, the mono-

tonicity of the payoff function in the private signal has to be shown.

20Continuity arguments show that such a simple switching strategy is optimal. Therefore,generality is not lost when imposing it in the first place.

21More precisely, each investor is indifferent at the border of the high dominance region,given that all other investors invest P (θ, 1) = 0 or at the border of the dominance region atthe high end of the support, if no one else invests P (θ, 0) = 0.


Lemma 3.1 P (θi, IT ) is strictly monotonically decreasing in the private signal

θi.


Due to the strict monotonicity of the payoff, the lowest possible threshold

for a switching strategy of all the investors is θ. Similarly, the highest possible

threshold is θ. For all θi < θ, the payoff is positive, irrespective of the actions

of all other investors. As the rationality of the investors is common knowledge,

not to invest is a dominated strategy for signals below θ. At the other extreme,

for all signals θi > θ, the payoff difference is negative.

Due to the strategic complementarity between investors, the worst scenario

that an investor must consider is the case where IT = Iθ. This case means that

for all values of the fundamentals in the multiplicity range, investors choose

not to invest although the fundamentals would, in case of coordination on the

high growth equilibrium, also allow for this. The best scenario would be a

switching strategy of IT = Iθ.

At this point, it is possible to start the iterated elimination of dominated

strategies. This iteration permits cutting the multiplicity range down to a

unique threshold signal. The elimination functions as follows: If an investor i

receives a signal that is very close to the border of the dominance region, the

probability that other investors receive signals within the dominance region

and, thus, have a dominant strategy is very high. Due to the strict mono-

tonicity, this suffices to induce the investor i to have a dominant strategy as

well. This is true for all the investors. Therefore, the range between the signal

of investor i and the former border of the dominance region can be added to

the dominance region. Performing this addition at both ends of the support

and iterating this process leads to the maximum [minimum] signal at which

investor i is indifferent between investing and not investing; this signal has

to be, at the same time, the threshold of the switching strategy of all other

investors.22

According to Milgrom and Roberts (1990), in all games with strategic

complementarity the set of strategies that resist the iterative elimination of

dominated strategies are limited by Nash equilibria. Nash equilibria are not

eliminated through this process. Thus Iθ? and Iθ? are the most extreme Nash

equilibria of the game. No Nash equilibrium exists below θ? in which the in-

22For a more formal consideration of the iterative elimination, please see Appendix 3.6.

3.2. THE MODEL 115

vestors do not invest. Likewise, no Nash equilibrium exists above θ?

in which

the investors invest.

Steps one and two enable the third step in the proof of the uniqueness of

the equilibrium. Given Lemma 3.1, it now suffices to show that equation

P (θi = θ?, Iθ?) = Rprob(θj ≤ θ?)− E(1

2θ2|θi) = 0 (3.6)

has a unique solution. This can be expressed in the following Lemma.

Lemma 3.2 There exists only one value, for which the expected payoff equals

0 given that investor i receives exactly the threshold signal θ? as a private

signal, and given that all other investors have a switching strategy, in which

the switching signal equals exactly θ?.


This unique solution is

θ? = (R− 1

3η2)

12 (3.7)

The three steps can be summarized in the following proposition:

Proposition 3.1 There exists a unique threshold equilibrium θ? of the game

with imperfect information, such that any investor i invests if and only if θi ≤θ? and does not invest if θi > θ?.

Proposition 3.1 permits the conclusion that θ? identified by Equation (3.7)

is the unique threshold equilibrium of the game with private information.

3.2.3 Results and Implications

The unique threshold equilibrium allows to show that an increase in the degree

of disagreement about the fundamentals in country B increases the probability

of a crisis there.

Proposition 3.2 A crisis becomes more likely to occur in country B if a sur-

prise crisis happens in country A. A crisis becomes less likely to occur in

country B if an anticipated crisis materializes in country A.


Proof. To prove Proposition 3.2, it suffices to calculate the comparative statics

of the unique threshold equilibrium in terms of the fundamentals with respect

to η.

These deliver the following result:

∂θ?

∂η=

1

2(R− 1

3η2)−

12 (−2

3η) < 0 (3.8)

This result implies that the threshold below which all investors invest shifts

to the left (right), i.e., to better (worse) levels of the fundamentals, if the

dispersion of private signals around the true value of the fundamentals increases

(decreases) due to a surprise (anticipated) crises in country A. Thereby, the

probability space of the good equilibrium is reduced (increased) and, hence, a

crisis becomes more (less) likely in the case where private signals are dispersed

in an η + c (η − d) surrounding of the true fundamentals, as opposed to the

case where they are only dispersed in an η surrounding.

Due to the assumption that higher uncertainty results from a crisis in

another country, e.g., Thailand in the case of Korea, the shift of the threshold

can be viewed as an incident of contagion.23

θB*‘

PB

PB (η+c)

θB*

with c>0

θB, θBi

PB (η)

θB*‘

PBPB

PB (η+c)

θB*

with c>0

θB, θBi

PB (η)

Figure 3.3: θ?B as a function of the dispersion of private signals in country B

Figure 3.3 illustrates the comparative static analysis. The payoff is plotted

against the level of the fundamentals. θ?′B lies at lower levels of the fundamentals

than θ?B as described above due to η + c being a higher value than η. Clearly,

the threshold based on a dispersion of private signals η− d would lie at higher

levels of the fundamentals than θ?B.

23This assumption is justified by empirical evidence, see Figure 3.1.


3.2.4 Testable Hypotheses

In this section, the predictions of the theoretical model are translated into

testable hypotheses. From Proposition 3.2, two testable hypotheses can be

derived:

Hypothesis 1 The occurrence of a surprise crisis in a first country makes a

crisis in a second country more likely through an increase in uncertainty about

the fundamentals in the second country.

Hypothesis 2 The occurrence of an anticipated crisis in a first country makes

a crisis in a second country more likely through a decrease in uncertainty about

the fundamentals in the second country.

3.3 Empirical Analysis

The purpose of this section is to validate the predictions of the theoretical

model. I focus on showing the effect of uncertainty about the fundamentals as

a channel through which crises spread from one financial market to another.

3.3.1 The Data

A rich data set is used comprising monthly observations of different alternative

crisis measures as the dependent variable, a measure of uncertainty as the main

explanatory variable, and a large set of control variables.24 The data run from

December 1993 to September 2005. The sample comprises 38 countries – 15

developed and 23 emerging – where the selection of the period and countries

reflects the existence of uncertainty and return data.25 I exclude the initial

crises countries (Argentina, Brazil, Mexico, Russia, Thailand, and Turkey)

from the set of potentially-affected countries. Although this means a non-

negligible loss in observations, this procedure is in favor of finding convincing

results.

The explanatory variable that is most interesting for the current analysis

is uncertainty about the fundamentals. As in the chapter on sudden stops

24Please refer to Table 3.2 in Appendix 3.6 for detailed descriptions of the time series andtheir calculation.

25For details, please refer to Table 3.1 in Appendix 3.6.


of capital flows, I use the standard deviation of growth forecasts by a group

of country experts as a measure of uncertainty. In models similar to Morris

and Shin (1998), uncertainty takes the form of the dispersion of private sig-

nals around the true value of the fundamentals. In the current model, this

is the dispersion of the private signals about the true value of the investment

environment. Such data is not directly observable. However, investment en-

vironments correlate strongly with the country levels of GDP and associated

growth. Hence, the available data by Consensus Economics on the standard

deviation of GDP growth forecasts between experts in an economy seems a

reasonable proxy.26

To measure the significant drops in stock returns, a crisis dummy variable

is constructed. Monthly stock market returns, computed from IFC (Interna-

tional Finance Corporation) investable US dollar total return index, serve as a

basis for this crisis dummy.27 When needed, I complete the returns with data

from MSCI (Morgan Stanley Capital International) or national sources.28 I

construct a binary crisis variable of severe drops in stock market returns, in

which a month is counted as a crisis month if the total return undershoots its

sample mean by more than two standard deviations. After this initial drop, the

subsequent months are also counted as crisis months until the return reverts

into the one standard deviation band around the sample mean.29

I use a rich set of domestic control variables. Most important are the

mean of the growth expectations by Consensus Economics to control for the

status of the economy and its evaluation by investors. Additionally, I disen-

tangle the effect of uncertainty about the fundamentals from effects linked to

the volatility of stock market returns, which I include as a control variable

into the regressions. Following Broner et al. (2006), I use the ICRG (In-

26Please refer to chapter 2 for detailed arguments why this measure is a good proxy ofthe dispersion of private signals around the true fundamentals of an economy. See Table3.2 in Appendix 3.6 for a description of the exact construction of the variable. In the mainanalysis, I use a weighted average of current and following year forecasts as described inTable 3.2. However, as a robustness check I repeat all estimations with the current year,and all estimations with following-year forecasts, separately. The results are qualitativelythe same and quantitatively similar.

27The investable indices take into consideration restrictions on foreign investment. There-fore, this measure represents the part of the national stock markets accessible to foreigninvestors, which is relevant in the context of contagion.

28For more details, please refer to Table 3.2 in Appendix 3.6.29I run the regressions with variants of this measure, i.e., 1.5 standard deviations and also

3.


ternational Country Risk Guide) indices of financial, economic, and political

risk as a summary statistics to control for the state of the fundamentals in

the potentially-affected country. Then, domestic liability dollarization, TOT

growth, and credit growth are included as further control variables.

Numerous alternative mechanisms of contagion appear to be relevant in the

context of stock market drops. Specifically, I control for contagion through

common creditors. In line with Van Rijckeghem and Weder (2001), I use

consolidated data of BIS banking statistics to construct an index of contagion

through the presence of a common creditor. However, I construct a different

index than their index. The index used in this chapter of the study reflects the

dependence on common creditors as opposed to their measure that reflects the

competition for funds. In the context of stock market drops, the dependence

appears more relevant than competition for their funding.30 Another relevant

channel of contagion is trade with the crisis country. Following Glick and Rose

(1999), I use bilateral export data from the IMF Direction of Trade Statistics

to construct the measure of trade contagion. However, in contrast to their

contagion measure, I use the export share to country A in total exports. For

the control of contagion through common overexposed fund investments, I

use the index developed by Broner et al. (2006). I interact the alternative

contagion measures with the crisis dummies for country A. It seems natural

that the contagion variables only play a role for the uncertainty in country B

if there is a crisis in country A to begin with.

3.3.2 Methodology

The goal of this study is to show that uncertainty about the fundamentals

has a separate and non-negligible effect on the spread of crises apart from the

channels already studied. If the goal of the present study were to prove the

relevance of uncertainty and its predominant role relative to other potential

explanatory variables in spreading financial crises, the best procedure to prove

this point would be a two-step instrumental variable estimation.31

In the context of analyzing contagion, it would be difficult to find a valid

instrument for the uncertainty in country B. Arguing, for example, for the use

30For detail on the construction of this index, please refer to Table 3.2 in Appendix 3.6.31A good example of a convincing instrumental variable estimation is Acemoglu, Johnson,

and Robinson (2001) who analyze institutions as opposed to geography as explanation ofdifferences in current dispersion of countries’ incomes.


of the crisis in country A as an instrument for the uncertainty in B requires the

crisis in A to significantly affect the uncertainty in B but not to directly affect

the probability of a crisis in country B and not to affect it through a channel

different from uncertainty. The existing literature on alternative channels of

contagion already proves the last assumption wrong. Other variables linked

to the crises in A, which might serve as instruments for the uncertainty in

B, would have the same problem: they are likely to also feed into alternative

channels of contagion.

Given the presence of alternative contagion channels other than the uncer-

tainty channel and therefore the impossibility of finding a valid instrument for

uncertainty in B, this empirical analysis is designed in the following way:32

In a first step, the effect of the crisis in country A on the uncertainty in a

second country B is estimated. To ensure that the effect of the crisis in A on

the uncertainty in B is correctly quantified, I control for potential domestic

drivers of uncertainty. I also control for country and time effects. Thereby, I

employ a very strict test on the effect of uncertainty on the occurrence of a

crisis. The control for time effects is often avoided in the literature. In the

second step, I analyze the effect of uncertainty in country B on the probability

of a crisis there. In this step, I run probit regressions estimating the effect of

the uncertainty in B on the probability of crises there. I control for domestic

factors that could trigger crises and also for alternative contagion channels.

Additionally, I control for country and time effects.

One drawback of this approach is that in contrast to an instrumental vari-

able estimation, reverse causality from the crisis in B on the uncertainty there

cannot be entirely ruled out. However, as described in more detail in subsec-

tion 3.3.3, I run a number of regressions to be confident that this possibility is

minimized in the chosen setup.

Methodology Used to Estimate the Effect of the Initial Crisis on theUncertainty in Potentially-affected Countries

To analyze the relevance of the uncertainty channel of contagion, I proceed

in two distinct steps. In the first step, I pin down the effect of the crisis in

country A on the uncertainty in a second country B. In the second step, I

analyze the effect of uncertainty in country B on the probability of a crisis

32In section 3.3.3, the reasons for this design are described in further detail.


there.

In step one, I estimate two sets of regressions. Firstly, I specify the following

test:

uncB,t = α0

+ α1d crArg,t−1 + ... + α6d crTur,t−1

+ α7d crB,t−1 + α8macrocntrlsB,t + δB + εB,t (3.9)

with B = 1, 2, ..., 32; t = 1, 2, ..., 141,

where uncB,t signifies the uncertainty in the potentially-affected country

B at time t. I exclude the initial crises countries Argentina, Brazil, Mexico,

Russia, Thailand, and Turkey in the panel as potentially-affected countries.

Therefore, the index B represents the 32 remaining countries in the sample.

d crA,t−1 signifies the lag of the crisis dummies in the initial-crisis countries A,

representing Argentina, Brazil, Mexico, Russia, Thailand, and Turkey. The

dummy variable takes a value of 1 if there is a significant drop in the stock

market return. macrocntrlsB,t−1 stands for the set of domestic control variables

described in section 3.3.1. δB stands for country specific effects. The level of

uncertainty varies strongly across countries.33 Systematically, some countries

are characterized by higher uncertainty than other countries, therefore, requir-

ing control for country effects. I run fixed-effects regressions to accommodate

this fact. Finally, εB,t stands for the error term.

Controlling for time effects in the above setting is not possible because

the average effect of each of the initial crises on the uncertainty in all the

countries contained in the sample is estimated. As the coefficients for each

of the crises are forced to be the same in the regression in all the potentially-

affected countries, it could be that the coefficients of the time dummies capture

part of the effect that actually comes from the crisis variable. To circumvent

this problem, I interact the crisis variable with the distance between the crisis

variable and the potentially-affected countries. I employ the distance variable

first used by Rose (2004). This creates heterogeneity in the crisis variable

across countries, which is necessary to be able to control for time effects.

33This is illustrated in chapter 2, in which I first introduce the measure of uncertainty.


Therefore, I run additional regressions based on the following equation:

uncB,t = α0

+ α1d crArg,t−1disArg,B + ... + α6d crTur,t−1disTur,B

+ α7d crB,t−1 + α8macrocntrlsB,t + δB + γt + εB,t (3.10)

where all the abbreviations have the same meaning as in Equation (3.9).

Additionally, the terms disArg,B and γt stand for the distance from initial-crisis

country to the potentially-affected country B and for the time effects, respec-

tively. To estimate Equation (3.10), I also run fixed-effects panel regressions,

additionally controlling for time effects.

Methodology Used to Estimate the Effect of the Uncertainty on theProbability of a Crisis in Potentially-affected Countries

For the analysis of the effect of uncertainty on the probability of a crisis in a

potentially-affected country, I specify the following estimation equation:

Prob(d crB = 1|xB,tβ0) = G(β0

+ β1

∑A=Arg,...,Tur

(ctgA,B,t−1 ∗ d crA,t−1)

+ β2uncB,t−1 ∗∑

A=Arg,...,Tur

(d crA,t−1) + δB + γt + εB,t)

(3.11)

with B = 1, 2, ..., 32; t = 1, 2, ..., 141.

Since this study is interested in the increase of the probability of a crisis,

I employ probit estimations. Hence, G(.) is the standard normal cumulative

distribution function. ctgA,B,t−1 represents the alternative channels of conta-

gion from country A to country B: common creditors, trade, dependence on

a common overexposed fund investor, and finally also the market size of the

crisis country. These are interacted with the crises in the initial-crisis countries

taking into account that it is important to control for their effect in transmit-

ting those crises to country B. I also include the interaction of uncertainty

with the crises because it is the effect of uncertainty – if a crisis in country A

takes place – which is of interest.

Probit models do not lend themselves to consistent estimates of the co-

efficients in a fixed effects regression. Hence, instead of a fixed-effects panel


estimation, a pooled probit introducing dummies can capture the country ef-

fects and time effects. Additionally, I estimate a linear probability model to

overcome the potential incidental-parameter problem that can arise in the de-

scribed procedure. As a plausibility check, I repeat the regressions with the

continuous return as dependent variable and run simple OLS regressions.

Using the interaction term of the uncertainty measure in B with the sum

of crises variables in countries A as a regressor implies the following risk: It

could be that the coefficient on this term simply picks up the direct effect of

the crises in A on a crisis in B. To ensure that this is not the case, I estimate

a set of regressions, in which I enter the uncertainty variable and the initial

crises variables separately. In these very simple regressions, I use the following

specification:

Prob(d crB = 1|xB,tβ0) = G(β0

+ β1d crA,t−1 + ... + β6d crA,t−1

+ β7uncB,t−1 + β8mgexpB,t−1 + δB + εB,t)

(3.12)

with B = 1, 2, ..., 32; t = 1, 2, ..., 141.

Again, the abbreviations stand for the same variables as before. Further-

more, I put the mean growth expectations explicitly in Equation (3.12) to

emphasize that it is used in this simple regression as a summary of the situa-

tion of country B.

3.3.3 Results

The results of the empirical analysis suggest that the uncertainty channel of

contagion does play a role in spreading crises across markets. First, I find a sig-

nificant and robust effect of the initial crisis on the uncertainty in potentially-

affected countries. Second, I find a significant and robust effect of the uncer-

tainty in the second country on the probability of a crisis there.34

34In the following, I show the results calculating the stock market returns from the MSCIindex, using a return drop of more than 2 standard deviations below the sample meanas crisis criterion and employing a weighted average of current and following year GDPforecasts as basis for the uncertainty measure. However, I have run the estimations alsowith the return data from IFC, with two variations of the crisis criterion, and with thecurrent year and the following year forecasts separately. The results of these different sets of


The Effect of a Crisis in an Initial-crisis Country on the Uncertaintyin a Potentially-affected Country

The analysis of the effect of an initial crisis on disagreement about the funda-

mentals in potentially-affected countries shows an interesting pattern: I find

that the Mexican, Russian, and Thai crises significantly increase disagreement

about the fundamentals in other countries. The literature identifies these crises

as surprise crises.

However, in the case of the three other crises in the sample – the Brazilian,

Turkish, and Argentinean crises – the panel analysis shows a different pattern:

The Turkish and Argentinean crises significantly decrease the uncertainty in

potentially-affected countries. The effect of the Brazilian crisis is less clear.

The literature identifies these crises as anticipated crises.

These results are robust to choosing regional sub-samples, emerging mar-

kets sub-samples, and including a large number of control variables. Table

3.5 summarizes the results of the fixed-effects panel regressions with those

sub-samples.

(Table 3.5 here)

The different columns in Table 3.5 correspond to the regression results from

different sub-samples. Column 1 shows the coefficients of the fixed-effects panel

regression of uncertainty in all potentially-affected countries on the crises in

all the initial crises countries and a set of control variables. Column 2 displays

regression results of the regression of the crises in all initial crises countries

on uncertainty in emerging market economies. Columns 3 to 10, then, show

results for regressions of the regional crises in the sub-samples of all economies

(columns 3, 5, 7, 9) and only the emerging economies (columns 4, 6, 8 and 10)

within Asia (columns 3 to 6), within Eastern Europe (columns 7 and 8) and

within Latin America (columns 9 and 10).

In all regressions, the lag of the 1994 Mexican, the 1998 Russian, and the

1997 Thai crises have a significant and positive effect on the uncertainty in the

analyses are qualitatively the same and quantitatively similar. Therefore, I do not includethem in this chapter.


potentially-affected countries. Comparing columns 9 and 10 with columns 1

and 2 reveals that the effect of the Mexican crisis is stronger on uncertainty in

Latin American countries (they are all emerging countries, which explains why

columns 9 and 10 are identical) than on the entire sample of countries or the

sub-sample of all emerging economies. A crisis event in Mexico leads to an in-

crease of the standard deviation of growth expectations across country experts

of 0.174 percentage points in Latin American countries. The Mexican crisis

exerts a smaller effect on uncertainty in the sample of all emerging markets,

increasing the standard deviation of growth expectations by 0.059 percentage

points. This effect is slightly bigger than the one observed in the sample of all

countries, where the increase is 0.044 percentage points.

The same pattern holds for the Russian and Thai crises. However, for these

two crises, the difference in the magnitude of the effect within their own region,

compared to the effect on the entire sample of countries, is not as large as for

the Mexican crisis. The effect of the Thai crisis on uncertainty in emerging

Asia is an increase of 0.123 percentage points of the standard deviation, while

its effect on all Asian countries is a bit smaller: 0.112. In the sample of all

emerging markets, the effect of the Thai crisis is 0.085 percentage points and

in the sample of all the countries, the effect is 0.063. While the Russian crisis

increases the standard deviation of growth expectations in Eastern European

countries by 0.145 percentage points, its effect on all emerging and all countries

amounts to 0.086 and 0.048 percentage points only.

These results suggest that the Mexican crisis has the strongest effect on

uncertainty in other countries, in magnitude within its own region among the

three mentioned crises. However, the Mexican crisis has less impact beyond

its own region than have the Russian and the Thai crises. Furthermore, these

results suggest that the Thai crisis has the biggest effect of all three crises in

the developed world.

A closer look on results for the 2002 Argentinean, the 1999 Brazilian, and

the 2001 Turkish crisis reveals a different picture. While the Argentinean crisis

decreases the standard deviation of growth expectation by 0.146 percentage

points in Latin American countries, the effect is weaker in the sample of all

emerging markets and all countries: a decrease of 0.039 and 0.026, respectively.

In case of the Brazilian crisis, only the decrease of 0.027 percentage points of

standard deviation of the growth expectation in the sample of all countries is

significant at the five-percent level, while the effect of this crisis is insignifi-


cant in all the sub-samples. The Turkish crisis delivers the same pattern as

the Argentinean crisis. However, the Turkish crisis presents one interesting

additional finding. First, estimating the effect of the Turkish crisis in the sub-

sample of Eastern European countries shows that the Turkey crisis yields a

decrease of uncertainty of 0.043 measured in the standard deviation of growth

expectations in those countries. Second, estimating the effect of the Turkish

crisis in the subsamples of emerging Asian countries and all Asian countries,

the effect is much stronger: There the crisis in Turkey results in a decrease of

0.1 and 0.074 percentage points respectively.

These results suggest that the negative effects of the Argentinean and Turk-

ish crises on uncertainty in potentially-affected countries are stronger within

their own regions than beyond. The Turkish crisis shows a bigger effect in Asia

than in Eastern Europe. The effect of the Brazilian crisis is less clear.

The coefficients on the control variables used in the regressions have the

expected signs. In particular, as expected, the lag of the mean of the growth

expectations impacts uncertainty negatively. This variable can be seen as a

summary of the state of the fundamentals and the expectations about it. If

the fundamentals are good – or everyone expects them to be good – then

disagreement about the fundamentals decreases. The lag of the crises in the

potentially-affected countries shows a positive and significant effect on uncer-

tainty. Past stock market volatility also has a strong positive and significant

effect on uncertainty.35 Additionally, I use the ICRG financial, economic, and

political risk indices as summary of the fundamentals following Broner et al.

(2006). The coefficients on these variables are mostly not significant in the

regressions.

To further ensure the robustness of the effects that the above regressions

reveal, I run a second set of regressions, controlling for time effects. As ex-

plained in section 3.3.2, controlling for time effects in the above setting is not

possible. To circumvent this problem, I interact the crisis variable with the

distance between the crisis variable and the potentially-affected countries.36

35By introducing the stock market volatility, I lose India from the sample and also lose anon-negligible amount of observations. Therefore, I have run all the regressions also withoutthe stock market variable. The results are qualitatively and quantitatively very similar. Forthis reason only the results including the stock market variable are displayed. In regressionswithout the stock market volatility, the coefficient on past crisis in country B is slightlyhigher.

36By using the distance variable, I lose Slovakia and Taiwan, for which the distances to


This creates heterogeneity in the crisis variable across countries, making con-

trol of time effects possible. Clearly, the meaning of the explanatory variable

is slightly changed. Now additionally, whether the distance in the sense of

Rose (2004) increases or decreases the effect of a crisis on the uncertainty in

the potentially-affected countries, makes a difference.

I repeat the above fixed-effects panel regression, replacing the lagged crises

variables for the Argentinean, Brazilian, Mexican, Russian, Thai, and Turkish

crisis with the interaction term between those crises variables with the distance

to the potentially-affected countries and control for time effects in addition.

Table 3.6 in Appendix 3.6 displays the results. The overall pattern of effects

remains the same as in the first set of regressions. The effects are still highly

significant. The only exception is the coefficient on the interaction term of the

Russian crisis with the distance variable in the regressions with the samples of

all and of all emerging markets, which become insignificant.

As the effect of the crises in the initial countries on uncertainty becomes

smaller, the question arises whether this stems from the interaction of the crises

with the distance or from the control for time effects. To disentangle these two

cases, I also run regressions with the interaction variables without controlling

for time effects. The results are displayed in Table 3.7 in Appendix 3.6. The

regression results are very similar to the ones where I control for time effects.

This result suggests that the interaction with the distance variable explains

the lower coefficients and thus the weaker effect of the crisis in the initial-crisis

countries on uncertainty; the control for time effects is not driving this result.

Hence, it is safe to say that the effect of the initial-crisis country diminishes

with an increasing distance. Taken together, these regression outcomes confirm

the observations from the first set of regressions. The results are robust against

the inclusion of time effects.

To summarize the findings of the first step of the analysis: The analysis

shows that the Mexican, the Russian, and the Thai crises significantly increase

uncertainty in potentially-affected countries. The effect is stronger within the

region where the crisis takes place. The Argentinean, the Turkish, and, to a

lesser extent, the Brazilian crises decrease uncertainty in potentially-affected

countries. These last three crises have a stronger negative effect within their

region. The effect appears to decrease with increasing distance.

the initial crises countries are not available in the data set underlying Rose (2004).


These findings are in line with the hypotheses derived from the theoretical

model. Recall that surprise crises increase uncertainty in other countries, while

anticipated crises decrease uncertainty in other countries. The findings regard-

ing the different regional effects are not captured by the theoretical model.

The Effect of the Uncertainty on the Probability of a Crisis in aPotentially-affected Country

In the second step of the analysis, I show robustly that uncertainty in the

potentially-affected country increases the probability of a crisis there. These

results are summarized in Tables 3.8 to 3.12.

Firstly, I run a pooled probit regression of the crises in the potentially-

affected countries on the interaction of uncertainty in country B, with the

sum of all initial crises countries, controlling for a set of variables including

country and time effects. Apart from the controls for country and time effects,

these variables classify in two categories: 1) domestic control variables and

2) alternative contagion channels, which could influence the likelihood of a

crisis in the potentially-affected countries. Table 3.8 displays the results. The

results of the pooled probit estimations including country and time controls

are displayed in column 1 for the sample of all countries and column 2 for the

sample of all emerging market countries. The estimation results of the linear

probability model are shown in columns 3 and 4. The outcomes of the simple

OLS regressions with the continuous return variable as dependent variable

appear in columns 5 and 6.

Columns 1 and 2 of Table 3.8 show the lag of uncertainty interacted with

the sum of crises in all initial crises countries has a positive and strongly sig-

nificant effect on the probability of a crisis in the potentially-affected country.

The effect is stronger in emerging economies. Introducing the dummies into

the pooled probit regression does not seem to create a severe incidental pa-

rameter problem. The magnitude of the effect is, indeed, smaller in columns 3

and 4 but the effect is still strongly significant and not negligible. The effect

of the uncertainty on the continuous return variable in columns 5 and 6 not

being significant is not problematic. The theoretical model is about crises,

which are extreme events. The regression with the continuous return variable

as regressand is a plausibility check, only. For example, if an increase in un-

certainty increased the return, while simultaneously increasing the probability


of a crisis, this would worry.

Secondly, I run the regressions with the interaction of uncertainty with

the sum of the crises in Mexico, Russia and Thailand. These are the crises

identified to increase the uncertainty in other countries. The results of these

regressions are summarized in Table 3.9.

(Table 3.9 here)

Clearly, the effects of uncertainty are stronger in the current regression

than in those with all initial crises countries.37 Here, also the coefficients of

the uncertainty term are significant when using the continuous return variable

as regressand.

Calculating marginal effects makes clear that the effect of the uncertainty

on the probability of crises in the potentially-affected countries is not negligible.

Details are shown in Table 3.12.

(Table 3.12 here)

The control variables have the expected signs. With regard to country

characteristics the following variables are controlled for: the lag of the mean

growth expectations, the lag of stock market volatility, and the ICRG risk

indices for economic, financial, and political risk.

With regard to alternative contagion channels, the following variables are

controlled for: the common creditor channel of contagion, the direct trade

channel, contagion from important stock markets, and contagion through com-

mon overexposed fund investors. I use slightly different definitions than the

37This finding goes beyond what is explained by the theoretical model, which would notdistinguish the intensity of an increase or decrease of uncertainty after a surprise crisis oran anticipated crisis.


literature to construct the index of common creditors and the index of trade

share with the initial-crisis country. The definitions that I use are more plausi-

ble in the context of stock market drops, rather than the existing indices which

have been developed to study contagion of currency crises. Section 3.3.1 ex-

plains the construction of these variables exactly. I estimate regressions, which

include the channel through overexposed fund investors separately, and show

the results in Table 3.10. This is due to the fact that I have the index of over-

exposed common creditors only for the sample of emerging markets without

Ukraine.38

The market size of the initial-crisis country turns out to be not significant

in the regressions (see Tables 3.8 to 3.10). If not controlling for overexposed

common fund investors, I find that common creditors and trade share have a

high explanatory power for the occurrence of a crises in the potentially-affected

countries (see Tables 3.8 and 3.9). However, if I introduce the control of the

overexposed common fund investor index, these two variables become insignif-

icant, which makes them appear not entirely robust, at least in the emerging

market sample, for which I can test the overexposure channel. Notably, the

overexposure channel cannot significantly contribute to the explanation of the

continuous fund returns in columns 5 and 6 in Table 3.10, while Broner et al.

(2006) find a significant effect. This difference could stem from the severe test

with control for country and time effects that I run in the present analysis.

As explained in section 3.3.2, care must be taken to ensure against only

picking up the direct effect of the crises in the initial-crises countries when

interacting the uncertainty in B with the sum of crises in the initial-crises

countries. The results in Table 3.11 show that the uncertainty has a distinct

positive and strongly significant effect on a crisis event in the same economy.

The possibility of reverse causality is not tackled in this second step of my

empirical analysis. This problem could arise if the crisis in B itself caused

the uncertainty to increase. There are two answers to this concern: First,

the present analysis is interested in the uncertainty that is caused by crises

elsewhere. I show robustly that a crisis in A significantly influences uncertainty

in country B. In this step, reverse causality is unlikely. Hence, this part of the

analysis is not affected by the endogeneity concern.

38I am very thankful to Broner et al. (2006) for making their overexposure index availableto me. Due to the expensive underlying source data, I would not have been able to controlfor this relevant contagion channel otherwise.


The problem arises only in the second step. Here, singling out perfectly

the uncertainty caused by the crisis in country A is not possible. However,

interacting uncertainty in B with the crises in the initial crises countries and,

at the same time, controlling for domestic causes of increased uncertainty pro-

vides the second answer to the endogeneity concern. The interaction allows to

consider exclusively the relevant time periods. Therefore, uncertainty caused

by the crises in the other countries should be especially high. Additionally,

controlling for the fundamentals in country B in the same regressions corrects

for domestic causes of increased uncertainty in B. As a check in the second

step, I estimate a set of regressions, in which I include uncertainty and crisis

indices for the initial crises countries both as separate explanatory variables.

Also in these regressions, the effect of uncertainty is positive and strongly sig-

nificant. In a follow-up check, I run a further set of regressions, including

exclusively either the crises indices in countries A or the uncertainty in B as

explanatory variables. The coefficient of uncertainty does not change signifi-

cantly. Thereby, I make sure that the uncertainty has a separate effect on the

probability of crises in B from the direct effect of the crisis in A.

Instrumental variable estimation is not an answer to the endogeneity con-

cern in the present setup. Instrumenting uncertainty, for example, by its past

realizations does not help. In the case where the instrument reaches far enough

back into the past, the instrument might be realized before the crises in the

initial-crisis country. It would then pick up exactly the part of uncertainty that

is not of interest in this analysis. Therefore, the series of checks conducted in

the present study appear to be the best available option.

These arguments support plausibility and reliability of the results. Further

soundness comes when taking these arguments together with the results of the

analysis on sudden stops in chapter 2. There, I show the strongly significant

effect of uncertainty on the occurrence of a crisis after taking care of the

endogeneity problem.

Together with the findings in section 3.3.3, the sets of regressions in the

current section support the theoretical model. Uncertainty in potentially-

affected countries increases with the occurrence of a surprise crisis in initial-

crises countries. In turn, this increase in uncertainty leads to an increase of the

probability of a crisis in potentially-affected countries. In case of an anticipated

crisis in the first country, the uncertainty in the second country is reduced. In

turn, this decreased uncertainty decreases the probability of a crisis in the


second country. The fact that the coefficients on the interaction term in Table

3.8 are larger than those in Table 3.9 suggests that the decreasing effect of

a decreasing uncertainty on the crisis probability after an anticipated initial

crisis is weaker than the increasing effect of an increased uncertainty after a

surprise crisis in the initial-crisis country on the crisis probability. However,

the weaker results in Table 3.8 could also stem from the Brazilian crisis not

having a clear-cut effect on the uncertainty in potentially-affected countries.

These empirical results align with the theoretical model. First of all, a large

part of the literature agrees that there was much less international response in

form of crisis in other countries to the Brazilian, the Turkish, and the Argen-

tinean crisis than to the three other crises.39 Additionally, Didier et al. (2006)

and Mondria (2006a) argue that the Brazilian, Turkish, and Argentinean crises

were anticipated by the investors while the Mexican, Thai, and Russian crises

caught them by surprise.

3.4 Policy Implications

The first, obvious implication of my analysis is that investors and governments

should closely monitor fundamentals also of other countries, especially in the

region and in adjacent countries. Surprise crises appear to be especially bad be-

cause they set off mechanisms that worsen the situation further. This chapter

illustrates such a mechanism through the uncertainty about the fundamentals.

Second, once a surprise crisis has hit a first country, governments need to

apply policies that counteract the increase in uncertainty about the funda-

mentals in country B. One venue could be to develop mechanisms for such

situations through which governments could disseminate credibly very precise

information about the state of their economy. In this model, I have not been

concerned with credibility issues, so I can only infer something about a cred-

ible government. In reality, governments might not be credible – they might

be tempted strongly to signal that the fundamentals in their country are very

satisfying. However, one way toward overcoming the credibility problem and

helping private investors receive more precise private signals, would be to al-

low full access to the government accounts to a few independent institutions,

which could then sell the information to private investors. Such a procedure

39See, for example, Kaminsky et al. (2000) or Didier et al. (2006).

3.5. CONCLUSION 133

ensures that private investors have private information but with little disper-

sion around the true value of the fundamentals. Another venue would be to

think about subsidies for information-gathering technology.

3.5 Conclusion

In this chapter, I illustrate the uncertainty channel of contagion in a coordina-

tion game and then validate the predictions empirically. In particular, I find

that surprise crises in an initial-crisis country such as the Mexican, Thai, or

Russian crises increase the probability of a crisis in other countries. In the case

of an anticipated crisis such as the Brazilian, Turkey, or Argentinean crises,

uncertainty is reduced making crises in potentially-affected countries less likely.

Additionally, the empirical analysis shows that the effects through uncertainty

are stronger in potentially-affected countries within the same region as and

closer to the initial-crisis country.

The results of the present analysis suggest that investors and governments

should closely monitor the fundamentals of neighboring countries to minimize

the risk of a surprise crisis. Second, policy makers should take uncertainty

about the fundamentals into account. Once a surprise crisis happens elsewhere,

policy makers should be ready to counteract the increase in the disagreement

about the fundamentals by adequate policies. Strategies that help private

investors receive precise private signals appear prudent in the light of this

analysis.

The present analysis also confirms the findings of the relevance of other

contagion channels especially through overexposed fund investors, also through

trade links and common creditors. However, on top of these channels, which

have been analyzed by the literature for some time, uncertainty does play a

role in explaining contagion patterns. And, as the analysis of marginal effects

shows, the effect is not negligible.

In this chapter, I have taken the change in dispersion of the private signals

in the second economy as given if a crisis happens in the first country. A

worthwhile future research agenda is to explicitly model the optimal choice

of spending on information-gathering technology. This would result in an

endogenous change in dispersion of the private signals in the second market.

As to the empirical analysis, future research moving to higher frequency


data, if available, could be worthwhile. This step might allow the exploration

of more convincing ways of determining the direction of causality.

3.6. APPENDIX 135

3.6 Appendix

3.6.1 Lemma Proofs

Proof of Lemma 3.1 (Monotonicity of P )

Proof. The monotonicity of P in θi can be very easily shown: In

P (θi, IT ) = R · prob(θi ≤ T )− E(1

2θ2|θi)

R does not depend on θi. In addition,

∂prob(θi ≤ T )

∂θi=

0 if T < θi − 2η and T > θi + 2η< 0 if θi − 2η < T < θi + 2η

Therefore, the term R · prob(θi ≤ T ) is weakly decreasing in θi. The term

−∂E(12θ2|θi)

∂θi< 0

is strictly decreasing in θi. As a consequence, P (θi, IT ) is strictly monotonically

decreasing in θi.

Proof of Lemma 3.2 (Uniqueness of Equilibrium)

Proof. Equation (3.6) can be rewritten as follows:

P (θ?, Iθ?) = R · prob(θj ≤ θ?)− 1

2η

∫ θ?+η

θ?−η

1

2θ2dθ = 0 (3.13)

This leads to

P (θ?, Iθ?) =1

2R− 1

2η[1

6θ3]θ

?+ηθ?−η = 0 (3.14)

This equation defines θ? and can easily be rearranged to equation (3.7).

3.6.2 Iterated Elimination of Dominated Strategies

The elimination is started at the borders of the dominance ranges. Due to the

strict monotonicity in θi, there exist unambiguous signals θ1

< θ0

= θ and

θ1 > θ0 = θ, such that

P (θi, Iθ0) < 0 for all θi > θ

1and P (θi, Iθ0) > 0 for all θi < θ1


As θ0

> θ0, it also holds that θ1

> θ1. For the case of the upper border of

the multiplicity area, this means: Given that the other agents do not invest

when receiving signals above θ0, the investment does not pay for signals above

θ1

either. Where I find θ1

by calculating P (θi = θ1, I

θ0). This process can

be iterated. Given that the other agents do not invest when receiving signals

above θn, it does not pay to invest at a signal θ

n+1with θ

n+1< θ

n. The

signals θn+1

are found by setting the expected payoff difference to 0, reflecting

indifference between investment and no investment for investor i:

P (θn+1

, Iθn) = R · prob(θj ≤ θ

n)− E(

1

2θ2|θi = θ

n+1) (3.15)

The sequence θn

is decreasing, monotone and bounded. By the common

knowledge of rationality, this process is driven to its limit of θ?

= limn→∞θn.

Concretely, it is possible to find a value θ?

such that

P (θ?, Iθ

?) = 0 (3.16)

θ?has the interpretation that above this signal all agents do not invest with

certainty.

At the lower bound of the multiplicity area, the analogue situation occurs,

just with the sequence θn being increasing. There one iterates until one finds:

P (θ?, Iθ?) = 0 (3.17)

That means, one iterates until one finds a maximum (minimum) signal at

which agent i is indifferent between investing and not, and which is at the

same time the threshold of the switching strategy of all other agents, when

starting off at θ0

= θ (θ0 = θ).

The switching strategies Iθ? and Iθ? are Nash equilibria of the private in-

formation game. According to Milgrom and Roberts (1990), in all games with

strategic complementarity, the highest and the lowest equilibrium that resist

the iterative elimination of dominated strategies are Nash equilibria. Put the

other way round: These Nash equilibria can never be eliminated. If θ?

= θ?,

there exists an unambiguous signal θ?, below which in equilibrium all agents

will invest and above which no one invests.

3.6. APPENDIX 137

3.6.3 Figures and Tables

Asia

&

Aus

tralia

East

ern

Euro

pe

Latin

Am

eric

a

Wes

tern

E

urop

e &

N

orth

A

mer

ica

Asia

&

Aus

tralia

East

ern

Euro

pe

Latin

Am

eric

a

Wes

tern

E

urop

e &

N

orth

A

mer

ica

Aus

tralia

Can

ada

Chi

naC

zech

Rep

ublicA

rgen

tina

Japa

nFr

ance

Hon

g K

ong

Hun

gary

Bra

zil

New

Zea

land

Ger

man

yIn

dia

Pola

ndC

hile

Italy

Indo

nesia

Rom

ania

Col

umbi

aN

ethe

rland

sK

orea

Rus

siaM

exic

oN

orw

ayM

alay

siaSl

ovak

iaPe

ruSp

ain

Sing

apor

eTu

rkey

Ven

ezue

laSw

eden

Taiw

anU

krai

neSw

itzer

land

Thai

land

UK

US

A

The

sam

ple

com

prise

s 38

cou

ntrie

s, o

f whi

ch 1

5 de

velo

ped

and

23 E

mer

ging

Mar

ket c

ount

ries.

Dev

elop

ed C

ount

ries

Em

ergi

ng M

arke

t Cou

ntrie

s

Ther

e ar

e 7

Latin

Am

eric

an, 1

2 A

sian

& A

ustra

lian,

8 E

aste

rn E

urop

ean

and

11 W

este

rn E

urop

ean

and

Nor

th

Am

eric

an c

ount

ries.

Table 3.1: Country sample


Var

iabl

esM

easu

res

Sou

rces

Cri

sis

mea

sure

:

Dro

ps in

sto

ck m

arke

t ret

urns

Bas

is fo

r the

cris

is m

easu

re is

the

perc

enta

ge c

hang

e in

US

$ na

tiona

l, to

tal r

etur

n st

ock

mar

ket i

ndic

es

rela

tive

to th

e pr

evio

us m

onth

. A m

onth

is c

ount

ed a

s a

cris

is m

onth

if th

e to

tal r

etur

n dr

ops

at le

ast 2

st

anda

rd d

evia

tions

bel

ow th

e sa

mpl

e m

ean.

S

ubse

quen

t mon

ths

are

also

cou

nted

as

crisi

s m

onth

s un

til th

e re

turn

mov

es b

ack

into

the

one

stan

dard

de

viat

ion

band

aro

und

the

sam

ple

mea

n.

2 se

ts o

f cris

is m

easu

res:

1)

IFC

inve

stab

le, t

otal

ret

urn

indi

ces,

2)

MS

CI

tota

l ret

urn

indi

ces.

In b

oth

case

s I

have

to u

se lo

cal s

ourc

es fo

r R

oman

ia, S

lova

kia

and

Rom

ania

. To

conv

ert t

hese

thre

e na

tiona

l ind

ices

from

na

tiona

l cur

renc

y to

US

$ I

use

exc

hang

e ra

te e

nd o

f pe

riod

data

from

the

IMF

: IF

S.

Exp

lana

tory

Var

iabl

e:

Unc

erta

inty

mea

sure

Mon

thly

dat

a of

the

wei

ghte

d av

erag

e of

the

stan

dard

de

viat

ions

of t

he c

urre

nt a

nd fo

llow

ing

year

fore

cast

of

eco

nom

ic g

row

th. T

he s

tand

ard

devi

atio

n is

ca

lcul

ated

for

the

expe

ctat

ions

that

exp

erts

for

the

spec

ific

econ

omy

utte

r at

a p

artic

ular

poi

nt in

tim

e. I

n Ja

nuar

y th

e st

anda

rd d

evia

tion

of th

e cu

rren

t yea

r fo

reca

sts

is w

eigh

ted

with

11/

12 a

nd th

e st

anda

rd

devi

atio

n of

the

follw

oing

yea

r fo

reca

sts

with

1/1

2. I

n F

ebru

ary

the

curr

ent y

ear

valu

e re

ceiv

es 1

0/12

w

eigh

t and

the

follw

ing

year

one

2/1

2. T

his

sche

me

cont

inue

s un

til D

ecem

ber

with

a w

eigh

ting

of 0

/12

for

the

curr

ent y

ear

valu

e an

d 12

/12

for

the

follo

win

g ye

ar v

alue

.

Con

sens

us E

cono

mic

s

Table 3.2: List of variables

3.6. APPENDIX 139

Var

iabl

esM

easu

res

Sou

rces

Dom

estic

con

trol

var

iabl

es

Mea

n gr

owth

exp

ecta

tions

Mon

thly

dat

a of

the

wei

ghte

d av

erag

e of

the

mea

n of

th

e cu

rren

t and

follo

win

g ye

ar fo

reca

st o

f GD

P

grow

th. T

he m

ean

is ca

lcul

ated

for

the

expe

ctat

ions

th

at e

xper

ts fo

r th

e sp

ecifi

c ec

onom

y ut

ter

at a

pa

rtic

ular

poi

nt in

tim

e. T

he w

eigh

ting

sche

me

is th

e sa

me

as fo

r th

e st

anda

rd d

evia

tion.

Con

sens

us E

cono

mic

s

Sto

ck m

arke

t vol

atilit

y S

tand

ard

devi

atio

n of

dai

ly r

etur

ns c

alcu

late

d fo

r a

time

win

dow

of t

hree

mon

ths

from

mon

th t-

2 to

t.C

f. cr

isis

mea

sure

. How

ever

res

tric

ted

avai

labi

lity

of

daily

retu

rns.

Cha

nges

in th

e po

litic

al ri

skD

iffer

ence

in p

oliti

cal r

isk

indi

catio

r re

lativ

e to

pr

evio

us m

onth

.IC

RG

Cha

nges

in th

e ec

onom

ic ri

skD

iffer

ence

in e

cono

mic

risk

indi

catio

r re

lativ

e to

pr

evio

us m

onth

.IC

RG

Cha

nges

in th

e fin

anci

al r

iskD

iffer

ence

in fi

nanc

ial r

isk in

dica

tior

rela

tive

to

prev

ious

mon

th.

ICR

G

Dis

tanc

e to

initi

al c

risis

coun

try

Cal

cula

ted

by R

ose

(200

4)R

ose

(200

4)

Dom

estic

liab

ility

dolla

rizat

ion

Dev

elop

ed c

ount

ries:

Loc

al a

sset

pos

ition

s in

fore

ign

curr

ency

by

BIS

rep

ortin

g ba

nks

as a

sha

re o

f GD

P.

EM

s: D

olla

r de

posi

ts p

lus

bank

fore

ign

borr

owin

g as

a

shar

e of

GD

P.

Cal

vo, I

zqui

erdo

, Mej

ia (

2004

): B

IS, H

onoh

an a

nd

Shi

(20

02),

Cen

tral

Ban

ks o

f Aus

tral

ia, N

ew Z

eala

nd,

Col

umbi

a, K

orea

, Bra

zil,

IMF:

IFS

TO

T g

row

th

Dev

elop

ed c

ount

ries:

Loc

al a

sset

pos

ition

s in

fore

ign

curr

ency

by

BIS

rep

ortin

g ba

nks

as a

sha

re o

f GD

P.

EM

s: D

olla

r de

posi

ts p

lus

bank

fore

ign

borr

owin

g as

a

shar

e of

GD

P.

Cal

vo, I

zqui

erdo

, Mej

ia (

2004

): B

IS, H

onoh

an a

nd

Shi

(20

02),

Cen

tral

Ban

ks o

f Aus

tral

ia, N

ew Z

eala

nd,

Col

umbi

a, K

orea

, Bra

zil,

IMF:

IFS

Cre

dit g

row

thC

redi

t to

priv

ate

seto

r, m

onth

ly r

ate

of c

hang

eIM

F: I

FS

Table 3.3: List of variables (continued)


Alte

rnat

ive

cont

agio

n ch

anne

ls

Inde

x of

com

mon

cre

dito

rs

Deg

ree

of d

epen

denc

e on

the

sam

e cr

edito

rs a

s th

e in

itial

cris

is co

untry

. I c

alcu

late

the

inde

x as

the

sum

ov

er a

ll cr

edito

rs o

f the

pro

duct

of t

he d

epen

denc

e of

co

untr

y B

on

cred

itor

i tim

es th

e im

port

ance

of t

he

initi

al c

risis

coun

try, A

, for

cre

dito

r i.

The

de

pend

ence

of c

ount

ry B

on

cred

itor

i is

the

ratio

of

the

volu

me

that

cou

ntry

B b

orro

ws

from

cre

dito

r i

rela

tive

to th

e to

tal b

orro

win

g by

cou

ntry

B. T

he

impo

rtan

ce o

f cou

ntry

A to

cre

dito

r i i

s th

e ra

tio o

f th

e vo

lum

e th

at c

redi

tor

i len

ds to

cou

ntry

A r

elat

ive

to to

tal l

endi

ng b

y cr

edito

r i.

BIS

qua

terly

rev

iew

, con

solid

ated

ban

king

dat

a, ta

ble

9B. I

con

stru

ct th

e m

onth

ly ti

mes

erie

s fr

om th

e qu

arte

rly o

bser

vatio

ns b

y in

terp

olat

ion.

Tra

de s

hare

Exp

orts

from

cou

ntry

B to

the

initi

al c

risis

cou

ntry

re

lativ

e to

tota

l exp

orts

by

coun

try

BIM

F: D

irect

ion

of tr

ade

stat

istic

s

Sto

ck m

arke

t siz

e M

arke

t val

ue o

f IF

C in

vest

able

/ F

TS

E in

dex

IFC

inve

stab

le/ F

TS

E in

dex

Inde

x of

com

mon

ove

rexp

osed

fu

nd in

vest

ors

Deg

ree

of d

epen

denc

e on

fund

inve

stor

s th

at a

re

over

expo

sed

to th

e in

itial

cris

is c

ount

ryB

rone

r, G

elos

, Rei

nhar

t (20

06)

Table 3.4: List of variables (continued)

3.6. APPENDIX 141

Pan

el r

egre

ssio

ns o

f u

ncer

tain

ty in

B w

ith

fixe

d ef

fect

s on

the

lag

of c

risi

s in

init

ial c

rise

s co

untr

ies

Ste

p 1

Effe

ct o

f cris

es in

the

initi

al c

rises

cou

ntrie

s on

the

unce

rtain

ty in

oth

er c

ount

ries

Dep

ende

nt V

aria

ble

unce

rtai

nty

in c

ount

ry B

at t

. The

initi

al c

rises

cou

ntrie

s ar

e ex

clud

ed a

s af

fect

ed c

ount

ries.

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

all c

rises

, all

coun

tries

all c

rises

, em

ergi

ng

mar

kets

regi

onal

cr

ises,

asia

regi

onal

cr

ises,

em

ergi

ng

asia

regi

onal

cr

ises

, asia

+

turk

ey

regi

onal

cr

ises

, em

ergi

ng

asia

+

turk

ey

regi

onal

cr

ises

, ea

ster

n eu

rope

regi

onal

cr

ises,

em

ergi

ng

east

ern

euro

pe

regi

onal

cr

ises,

latin

am

eric

a

regi

onal

cr

ises

, em

ergi

ng

latin

am

eric

a

-0.0

26**

-0.0

39**

-0.1

46**

*-0

.146

***

(0.0

11)

(0.0

19)

(0.0

41)

(0.0

41)

-0.0

27**

-0.0

21-0

.041

-0.0

41(0

.013

)(0

.024

)(0

.051

)(0

.051

)0.

044*

*0.

059*

0.17

4***

0.17

4***

(0.0

17)

(0.0

32)

(0.0

61)

(0.0

61)

0.04

8***

0.08

6***

0.14

5***

0.14

5***

(0.0

11)

(0.0

19)

(0.0

31)

(0.0

31)

0.06

3***

0.08

5***

0.11

2***

0.12

3***

0.10

8***

0.11

6***

(0.0

09)

(0.0

16)

(0.0

19)

(0.0

26)

(0.0

19)

(0.0

26)

-0.0

36**

*-0

.070

***

-0.0

74**

*-0

.100

***

-0.0

43*

-0.0

43*

(0.0

09)

(0.0

17)

(0.0

19)

(0.0

26)

(0.0

25)

(0.0

25)

0.16

4***

0.13

7***

0.22

7***

0.22

1***

0.23

0***

0.22

9***

0.06

4*0.

064*

0.13

6**

0.13

6**

(0.0

18)

(0.0

25)

(0.0

30)

(0.0

36)

(0.0

30)

(0.0

36)

(0.0

36)

(0.0

36)

(0.0

55)

(0.0

55)

-0.1

19**

*-0

.122

***

-0.1

22**

*-0

.122

***

-0.1

25**

*-0

.126

***

-0.1

03**

*-0

.103

***

-0.1

06**

*-0

.106

***

(0.0

02)

(0.0

03)

(0.0

04)

(0.0

05)

(0.0

04)

(0.0

05)

(0.0

08)

(0.0

08)

(0.0

09)

(0.0

09)

5.09

2***

5.46

8***

5.33

2***

5.76

4***

4.96

8***

5.19

0***

3.89

6***

3.89

6***

6.49

0***

6.49

0***

(0.3

57)

(0.4

93)

(0.7

34)

(0.8

70)

(0.7

34)

(0.8

73)

(0.7

23)

(0.7

23)

(1.3

55)

(1.3

55)

-0.0

000.

002

-0.0

04-0

.006

-0.0

03-0

.006

-0.0

02-0

.002

0.02

2*0.

022*

(0.0

04)

(0.0

06)

(0.0

06)

(0.0

08)

(0.0

06)

(0.0

08)

(0.0

13)

(0.0

13)

(0.0

11)

(0.0

11)

-0.0

07*

-0.0

14**

-0.0

09-0

.007

-0.0

09-0

.008

-0.0

05-0

.005

-0.0

21-0

.021

(0.0

03)

(0.0

06)

(0.0

07)

(0.0

09)

(0.0

07)

(0.0

09)

(0.0

10)

(0.0

10)

(0.0

13)

(0.0

13)

-0.0

03-0

.001

-0.0

11*

-0.0

22**

-0.0

10-0

.021

**-0

.003

-0.0

030.

012

0.01

2(0

.003

)(0

.005

)(0

.006

)(0

.009

)(0

.006

)(0

.009

)(0

.010

)(0

.010

)(0

.008

)(0

.008

)C

onst

ant

0.84

2***

1.10

0***

1.09

5***

1.23

2***

1.13

1***

1.28

9***

0.85

0***

0.85

0***

1.02

9***

1.02

9***

(0.0

12)

(0.0

22)

(0.0

30)

(0.0

44)

(0.0

31)

(0.0

46)

(0.0

27)

(0.0

27)

(0.0

46)

(0.0

46)

Obs

erva

tions

2657

1293

944

598

944

598

353

353

342

342

Num

ber o

f cou

ntry

3016

96

96

66

44

R-s

quar

ed0.

600.

650.

700.

740.

700.

740.

460.

460.

450.

45S

tand

ard

erro

rs in

par

enth

eses

, * s

igni

fican

t at 1

0%; *

* si

gnifi

cant

at 5

%; *

** s

igni

fican

t at 1

%

lag

of fi

nanc

ial r

isk in

dex

lag

of p

oliti

cal r

isk

inde

x

lag

of c

risis

in R

us

lag

of c

risis

in T

ha

lag

of c

risis

in T

ur

lag

of m

ean

gro

wth

ex

pect

atio

ns

lag

of c

risis

in B

lag

of s

tock

mar

ket

vola

tility

lag

of c

risis

in A

rg

lag

of c

risis

in B

ra

lag

of c

risis

in M

ex

lag

of e

cono

mic

risk

in

dex

Table 3.5: Step 1: Effect of crisis in A on uncertainty in B

142 CHAPTER 3. UNCERTAINTY AND SPREAD OF CRISESP

anel

reg

ress

ions

of

unc

erta

inty

in B

wit

h fi

xed

effe

cts

on t

he la

g of

cri

sis

in in

itia

l cri

ses

coun

trie

sSt

ep 1

Effe

ct o

f cris

es in

the

initi

al c

rises

cou

ntrie

s on

the

unce

rtai

nty

in o

ther

cou

ntrie

sD

epen

dent

Var

iabl

e un

cert

aint

y in

cou

ntry

B a

t t. T

he in

itial

cris

es c

ount

ries

are

excl

uded

as

affe

cted

cou

ntrie

s.(1

)(2

)(3

)(4

)(5

)(6

)(7

)(8

)(9

)(1

0)

all c

rises

, all

coun

tries

all c

rises

, em

ergi

ng

mar

kets

regi

onal

cr

ises,

asia

regi

onal

cr

ises,

em

ergi

ng

asia

regi

onal

cr

ises,

asia

+

turk

ey

regi

onal

cr

ises,

em

ergi

ng

asia

+

turk

ey

regi

onal

cr

ises,

ea

ster

n eu

rope

regi

onal

cr

ises,

em

ergi

ng

east

ern

euro

pe

regi

onal

cr

ises

, lat

in

amer

ica

regi

onal

cr

ises,

em

ergi

ng

latin

am

eric

a

-0.0

04**

*-0

.006

***

-0.0

16**

*-0

.016

***

(0.0

01)

(0.0

02)

(0.0

05)

(0.0

05)

-0.0

02-0

.001

-0.0

07-0

.007

(0.0

01)

(0.0

03)

(0.0

06)

(0.0

06)

0.00

9***

0.01

0**

0.02

8***

0.02

8***

(0.0

02)

(0.0

04)

(0.0

07)

(0.0

07)

0.00

00.

000

0.00

9**

0.00

9**

(0.0

00)

(0.0

01)

(0.0

04)

(0.0

04)

0.00

5***

0.00

9***

0.01

0***

0.00

9**

0.01

0***

0.01

0**

(0.0

01)

(0.0

02)

(0.0

03)

(0.0

04)

(0.0

03)

(0.0

04)

-0.0

02*

-0.0

03-0

.005

*-0

.005

0.00

30.

003

(0.0

01)

(0.0

02)

(0.0

03)

(0.0

04)

(0.0

03)

(0.0

03)

0.16

5***

0.15

3***

0.23

8***

0.21

2***

0.24

2***

0.21

8***

0.06

7***

0.06

7***

0.10

8**

0.10

8**

(0.0

17)

(0.0

25)

(0.0

32)

(0.0

39)

(0.0

32)

(0.0

39)

(0.0

25)

(0.0

25)

(0.0

49)

(0.0

49)

-0.1

21**

*-0

.132

***

-0.1

32**

*-0

.132

***

-0.1

33**

*-0

.133

***

-0.1

00**

*-0

.100

***

-0.1

21**

*-0

.121

***

(0.0

03)

(0.0

04)

(0.0

05)

(0.0

06)

(0.0

05)

(0.0

06)

(0.0

06)

(0.0

06)

(0.0

08)

(0.0

08)

4.90

4***

4.23

1***

3.79

9***

3.89

6***

3.73

2***

3.79

1***

0.72

00.

720

3.90

0***

3.90

0***

(0.4

25)

(0.6

56)

(0.7

86)

(0.9

74)

(0.7

85)

(0.9

76)

(1.0

51)

(1.0

51)

(1.2

27)

(1.2

27)

-0.0

03-0

.000

-0.0

05-0

.005

-0.0

04-0

.004

-0.0

18*

-0.0

18*

0.01

60.

016

(0.0

04)

(0.0

06)

(0.0

06)

(0.0

08)

(0.0

06)

(0.0

08)

(0.0

10)

(0.0

10)

(0.0

10)

(0.0

10)

-0.0

06*

-0.0

13**

-0.0

12*

-0.0

12-0

.012

*-0

.012

0.00

80.

008

-0.0

17-0

.017

(0.0

03)

(0.0

06)

(0.0

07)

(0.0

09)

(0.0

07)

(0.0

09)

(0.0

07)

(0.0

07)

(0.0

12)

(0.0

12)

-0.0

04-0

.003

-0.0

12*

-0.0

21**

-0.0

11*

-0.0

20**

-0.0

06-0

.006

0.00

80.

008

(0.0

03)

(0.0

05)

(0.0

07)

(0.0

09)

(0.0

07)

(0.0

09)

(0.0

08)

(0.0

08)

(0.0

07)

(0.0

07)

time

effe

cts

yes

yes

yes

yes

yes

yes

yes

yes

yes

yes

Con

stan

t1.

649*

**2.

312*

**2.

111*

**-0

.312

2.11

6***

-0.2

300.

588

0.58

82.

510*

**2.

510*

**(0

.108

)(0

.276

)(0

.235

)(0

.754

)(0

.235

)(0

.756

)(5

.086

)(5

.086

)(0

.352

)(0

.352

)O

bser

vatio

ns24

4010

7684

449

884

449

823

623

634

234

2N

umbe

r of c

ount

ry27

138

58

54

44

4R

-squ

ared

0.64

0.70

0.73

0.77

0.73

0.77

0.68

0.68

0.58

0.58

Stan

dard

err

ors

in p

aren

thes

es, *

sig

nific

ant a

t 10%

; **

sign

ifica

nt a

t 5%

; ***

sig

nific

ant a

t 1%

lag

of in

tera

ctio

n cr

isis

in

Arg

* di

stan

ce to

Bla

g of

inte

ract

ion

crisi

s in

B

ra*

dist

ance

to B

lag

of in

tera

ctio

n cr

isis

in

Mex

* di

stan

ce to

B

lag

of e

cono

mic

risk

inde

x

lag

of fi

nanc

ial r

isk in

dex

lag

of p

oliti

cal r

isk in

dex

lag

of in

tera

ctio

n cr

isis

in

Rus

* di

stan

ce to

Bla

g of

inte

ract

ion

crisi

s in

Th

a* d

istan

ce to

Bla

g of

inte

ract

ion

crisi

s in

Tu

r* d

istan

ce to

B

lag

of m

ean

gro

wth

ex

pect

atio

nsla

g of

sto

ck m

arke

t vo

latil

ity

lag

of c

risis

in B

Table 3.6: Step 1: Effect of crisis in A on uncertainty in B, interacted explana-tory variable, time effects

3.6. APPENDIX 143

Pan

el r

egre

ssio

ns o

f u

ncer

tain

ty in

B w

ith

fixe

d ef

fect

s on

the

lag

of c

risi

s in

init

ial c

rise

s co

untr

ies

Ste

p 1

Effe

ct o

f cris

es in

the

initi

al c

rises

cou

ntrie

s on

the

unce

rtain

ty in

oth

er c

ount

ries

Dep

ende

nt V

aria

ble

unce

rtai

nty

in c

ount

ry B

at t

. The

initi

al c

rises

cou

ntrie

s ar

e ex

clud

ed a

s af

fect

ed c

ount

ries.

Dat

a so

urce

: M

SC

I(1

)(2

)(3

)(4

)(5

)(6

)(7

)(8

)(9

)(1

0)

all c

rises

, all

coun

trie

s

all c

rises

, em

ergi

ng

mar

kets

regi

onal

cr

ises

, asi

a

regi

onal

cr

ises

, em

ergi

ng

asia

regi

onal

cr

ises,

asi

a +

turk

ey

regi

onal

cr

ises

, em

ergi

ng

asia

+

turk

ey

regi

onal

cr

ises

, ea

ster

n eu

rope

regi

onal

cr

ises,

em

ergi

ng

east

ern

euro

pe

regi

onal

cr

ises,

latin

am

eric

a

regi

onal

cr

ises

, em

ergi

ng

latin

am

eric

a

-0.0

05**

*-0

.008

***

-0.0

23**

*-0

.023

***

(0.0

01)

(0.0

02)

(0.0

06)

(0.0

06)

-0.0

05**

*-0

.007

**-0

.013

*-0

.013

*(0

.001

)(0

.003

)(0

.007

)(0

.007

)0.

012*

**0.

019*

**0.

041*

**0.

041*

**(0

.002

)(0

.004

)(0

.007

)(0

.007

)0.

001

0.00

00.

010*

**0.

010*

**(0

.000

)(0

.001

)(0

.003

)(0

.003

)0.

005*

**0.

008*

**0.

011*

**0.

014*

**0.

010*

**0.

013*

**(0

.001

)(0

.002

)(0

.003

)(0

.004

)(0

.003

)(0

.004

)-0

.005

***

-0.0

09**

*-0

.011

***

-0.0

15**

*-0

.003

-0.0

03(0

.001

)(0

.002

)(0

.002

)(0

.003

)(0

.003

)(0

.003

)0.

174*

**0.

165*

**0.

262*

**0.

259*

**0.

262*

**0.

263*

**0.

077*

**0.

077*

**0.

111*

*0.

111*

*(0

.018

)(0

.025

)(0

.033

)(0

.040

)(0

.032

)(0

.040

)(0

.026

)(0

.026

)(0

.053

)(0

.053

)-0

.120

***

-0.1

25**

*-0

.121

***

-0.1

21**

*-0

.126

***

-0.1

27**

*-0

.101

***

-0.1

01**

*-0

.108

***

-0.1

08**

*(0

.003

)(0

.004

)(0

.004

)(0

.006

)(0

.004

)(0

.006

)(0

.006

)(0

.006

)(0

.009

)(0

.009

)4.

983*

**5.

210*

**5.

436*

**5.

797*

**4.

830*

**4.

831*

**1.

969*

1.96

9*6.

633*

**6.

633*

**(0

.431

)(0

.661

)(0

.787

)(0

.961

)(0

.785

)(0

.968

)(1

.098

)(1

.098

)(1

.306

)(1

.306

)-0

.003

-0.0

00-0

.005

-0.0

07-0

.004

-0.0

04-0

.015

-0.0

150.

016

0.01

6(0

.004

)(0

.006

)(0

.007

)(0

.009

)(0

.007

)(0

.008

)(0

.010

)(0

.010

)(0

.011

)(0

.011

)-0

.006

*-0

.013

**-0

.011

-0.0

10-0

.011

-0.0

100.

007

0.00

7-0

.020

-0.0

20(0

.003

)(0

.006

)(0

.007

)(0

.009

)(0

.007

)(0

.009

)(0

.007

)(0

.007

)(0

.013

)(0

.013

)-0

.003

-0.0

01-0

.013

*-0

.025

**-0

.011

-0.0

22**

-0.0

10-0

.010

0.01

20.

012

(0.0

03)

(0.0

06)

(0.0

07)

(0.0

10)

(0.0

07)

(0.0

10)

(0.0

08)

(0.0

08)

(0.0

08)

(0.0

08)

time

effe

cts

nono

nono

nono

nono

nono

Con

stan

t0.

848*

**1.

164*

**1.

105*

**1.

276*

**1.

157*

**1.

361*

**0.

839*

**0.

839*

**1.

032*

**1.

032*

**(0

.014

)(0

.029

)(0

.031

)(0

.049

)(0

.032

)(0

.052

)(0

.031

)(0

.031

)(0

.045

)(0

.045

)O

bser

vatio

ns24

4010

7684

449

884

449

823

623

634

234

2N

umbe

r of c

ount

ry27

138

58

54

44

4R

-squ

ared

0.62

0.68

0.70

0.75

0.71

0.76

0.62

0.62

0.48

0.48

Sta

ndar

d er

rors

in p

aren

thes

es, *

sig

nific

ant a

t 10%

; **

sign

ifica

nt a

t 5%

; ***

sig

nific

ant a

t 1%

lag

of in

tera

ctio

n cr

isis

in

Arg

* di

stan

ce to

Bla

g of

inte

ract

ion

cris

is in

Bra

* di

stan

ce to

Bla

g of

inte

ract

ion

cris

is in

M

ex*

dist

ance

to B

lag

of e

cono

mic

ris

k in

dex

lag

of s

tock

mar

ket v

olat

ility

lag

of fi

nanc

ial r

isk in

dex

lag

of p

oliti

cal r

isk

inde

x

lag

of in

tera

ctio

n cr

isis

in

Rus

* di

stan

ce to

Bla

g of

inte

ract

ion

cris

is in

T

ha*

dist

ance

to B

lag

of in

tera

ctio

n cr

isis

in T

ur*

dist

ance

to B

lag

of m

ean

gro

wth

ex

pect

atio

ns

lag

of c

risis

in B

Table 3.7: Step 1: Effect of crisis in A on uncertainty in B, interacted explana-tory variable, no control for time effects


anel

reg

ress

ions

of

cris

is in

pot

entia

lly a

ffec

ted

coun

try

on t

he la

gged

unc

erta

inty

in B

Ste

p 2

Effe

ct o

f unc

erta

inty

in a

cou

ntry

on

the

prob

abili

ty o

f a c

risis

ther

e if

a cr

isis

take

s pl

ace

in a

n in

itial

cris

is c

ount

ryD

epen

dent

Var

iabl

e cr

isis

as m

easu

red

by s

igni

fican

t neg

ativ

e re

turn

s in

cou

ntry

B a

t tT

he c

onsi

dere

d cr

ises

per

iods

are

Mex

ico

1994

/5, T

haila

nd 1

997,

Rus

sia

1998

, Bra

zil 1

999,

Tur

key

2001

, Arg

entin

a 20

02

(1)

(2)

(3)

(4)

(5)

(6)

pool

ed p

robi

t all

cris

es, a

ll co

untr

ies

pool

ed p

robi

t all

cris

es, e

mer

ging

m

arke

ts

linea

r pro

babi

lity

all c

rises

, all

coun

trie

s

linea

r pro

babi

lity

all c

rises

, em

ergi

ng

mar

kets

OLS

, all

cris

es, a

ll co

untr

ies

OLS

, all

cris

es,

emer

ging

mar

kets

0.10

0**

0.12

5**

0.04

0***

0.02

5***

0.00

10.

003

(0.0

47)

(0.0

50)

(0.0

06)

(0.0

09)

(0.0

02)

(0.0

04)

5.82

5**

4.96

30.

165*

*1.

677*

*-0

.017

-0.3

16(2

.930

)(3

.475

)(0

.076

)(0

.653

)(0

.032

)(0

.267

)6.

918*

**7.

190*

**1.

441*

**1.

331*

**-0

.465

***

-0.4

75**

*(2

.551

)(2

.630

)(0

.264

)(0

.405

)(0

.113

)(0

.166

)-0

.000

-0.0

00-0

.000

*-0

.000

0.00

0***

0.00

0*(0

.000

)(0

.000

)(0

.000

)(0

.000

)(0

.000

)(0

.000

)0.

127*

**0.

120*

**0.

011*

*0.

010

-0.0

09**

*-0

.009

***

(0.0

41)

(0.0

44)

(0.0

04)

(0.0

07)

(0.0

02)

(0.0

03)

9.82

8**

7.31

01.

824*

**0.

981

-0.3

69*

-0.3

13(4

.350

)(4

.607

)(0

.488

)(0

.687

)(0

.208

)(0

.281

)-0

.064

-0.0

76-0

.007

-0.0

10-0

.001

-0.0

01(0

.045

)(0

.047

)(0

.005

)(0

.008

)(0

.002

)(0

.003

)-0

.043

-0.0

51-0

.009

**-0

.017

**0.

003

0.00

2(0

.038

)(0

.040

)(0

.004

)(0

.008

)(0

.002

)(0

.003

)-0

.019

-0.0

02-0

.005

-0.0

030.

001

0.00

1(0

.040

)(0

.042

)(0

.004

)(0

.007

)(0

.002

)(0

.003

)tim

e ef

fect

s ye

sye

sye

sye

sye

sye

sco

untr

y ef

fect

sye

sye

sye

sye

sye

sye

sC

onst

ant

-62.

138*

**-5

7.88

1**

-4.6

85**

*-8

.027

**0.

499

0.56

1(2

1.66

0)(2

5.94

1)(1

.533

)(3

.146

)(0

.654

)(1

.286

)O

bser

vatio

ns13

0990

319

2396

119

2396

1R

-squ

ared

0.24

0.26

0.04

0.05

Stan

dard

err

ors

in p

aren

thes

es, *

sig

nific

ant a

t 10%

; **

sign

ifica

nt a

t 5%

; ***

sig

nific

ant a

t 1%

lag

of p

oliti

cal r

isk

inde

x

lag

of e

cono

mic

risk

inde

x

lag

of fi

nanc

ial r

isk

inde

x

retu

rns

as d

ep. V

aria

ble

lag

of m

ean

grow

th

expe

ctat

ion

lag

of s

tock

mar

ket v

olat

iltiy

lag

of in

tera

ct. u

ncer

t. in

B &

cr

isis

in a

ll in

itial

cris

is c

ntrs

.la

g of

inte

ract

. com

cre

dito

rs

& c

risis

in A

lag

of in

tera

ct. t

rade

shar

e &

cr

isis

in A

lag

of in

tera

ct. m

arke

t siz

e &

cr

isis

in A

Table 3.8: Step 2: Effect of uncertainty in B on crisis there, all initial crisescountries

3.6. APPENDIX 145

Pan

el r

egre

ssio

ns o

f cr

isis

in p

oten

tially

aff

ecte

d co

untr

y on

the

lagg

ed u

ncer

tain

ty in

BSt

ep 2

Effe

ct o

f unc

erta

inty

in a

cou

ntry

on

the

prob

abilit

y of

a c

risis

ther

e if

a cr

isis

take

s pl

ace

in a

n in

itial

cris

is c

ount

ryD

epen

dent

Var

iabl

e cr

isis

as m

easu

red

by s

igni

fican

t neg

ativ

e re

turn

s in

cou

ntry

B a

t tTh

e co

nsid

ered

cris

es p

erio

ds a

re M

exic

o 19

94/5

, Tha

iland

199

7, R

ussia

199

8, B

razil

199

9, T

urke

y 20

01, A

rgen

tina

2002

(1)

(2)

(3)

(4)

(5)

(6)

pool

ed p

robi

t all

cris

es, a

ll co

untr

ies

pool

ed p

robi

t all

cris

es, e

mer

ging

m

arke

ts

linea

r pro

babi

lity

all c

rises

, all

coun

trie

s

linea

r pro

babi

lity

all c

rises

, em

ergi

ng

mar

kets

OLS

, all

cris

es, a

ll co

untr

ies

OLS

, all

cris

es,

emer

ging

mar

kets

0.28

6***

0.31

4***

0.09

2***

0.07

1***

-0.0

13**

*-0

.010

*(0

.078

)(0

.082

)(0

.010

)(0

.015

)(0

.004

)(0

.006

)5.

556*

5.13

60.

152*

*1.

605*

*-0

.003

-0.2

20(2

.865

)(3

.424

)(0

.075

)(0

.642

)(0

.032

)(0

.264

)5.

779*

*5.

867*

*1.

234*

**1.

132*

**-0

.380

***

-0.4

08**

(2.5

36)

(2.6

02)

(0.2

64)

(0.4

05)

(0.1

13)

(0.1

67)

-0.0

00-0

.000

-0.0

00-0

.000

0.00

0***

0.00

0***

(0.0

00)

(0.0

00)

(0.0

00)

(0.0

00)

(0.0

00)

(0.0

00)

0.14

9***

0.13

9***

0.01

1***

0.01

3**

-0.0

11**

*-0

.012

***

(0.0

40)

(0.0

43)

(0.0

04)

(0.0

06)

(0.0

02)

(0.0

03)

11.1

93**

8.91

8*2.

009*

**1.

179*

-0.3

43*

-0.2

91(4

.390

)(4

.615

)(0

.481

)(0

.678

)(0

.207

)(0

.279

)-0

.049

-0.0

61-0

.004

-0.0

07-0

.001

-0.0

01(0

.045

)(0

.047

)(0

.005

)(0

.008

)(0

.002

)(0

.003

)-0

.043

-0.0

52-0

.008

*-0

.017

**0.

002

0.00

2(0

.038

)(0

.040

)(0

.004

)(0

.008

)(0

.002

)(0

.003

)-0

.010

0.00

9-0

.004

-0.0

010.

001

0.00

1(0

.040

)(0

.042

)(0

.004

)(0

.007

)(0

.002

)(0

.003

)tim

e ef

fect

s ye

sye

sye

sye

sye

sye

sco

untr

y ef

fect

sye

sye

sye

sye

sye

sye

sC

onst

ant

-61.

549*

**-5

7.00

2**

-3.6

06**

-7.2

53**

0.52

10.

837

(21.

754)

(26.

069)

(1.5

09)

(3.0

79)

(0.6

49)

(1.2

67)

Obs

erva

tions

1309

903

1923

961

1923

961

R-s

quar

ed0.

250.

270.

040.

05St

anda

rd e

rror

s in

par

enth

eses

, * s

igni

fican

t at 1

0%; *

* si

gnifi

cant

at 5

%; *

** s

igni

fican

t at 1

%

lag

of p

oliti

cal r

isk

inde

x

lag

of e

cono

mic

risk

inde

x

lag

of fi

nanc

ial r

isk

inde

x

retu

rns

as d

ep. V

aria

ble

lag

of m

ean

grow

th

expe

ctat

ion

lag

of s

tock

mar

ket v

olat

ility

lag

of in

tera

ct. u

ncer

t. in

B &

cr

isis

in M

ex, R

us, T

hala

g of

inte

ract

. com

cre

dito

rs

& c

risis

in A

lag

of in

tera

ct. t

rade

shar

e &

cr

isis

in A

lag

of in

tera

ct. m

arke

t siz

e &

cr

isis

in A

Table 3.9: Step 2: Effect of uncertainty in B on crisis there, Mexican, Russianand Thai crises


anel

reg

ress

ions

of

cri

sis

in p

oten

tial

ly a

ffec

ted

coun

try

on t

he la

gged

unc

erta

inty

in B

Step

2 E

ffect

of u

ncer

tain

ty in

a c

ount

ry o

n th

e pr

obab

ility

of a

cris

is th

ere

if a

cris

is ta

kes

plac

e in

an

initi

al c

risis

coun

try

Dep

ende

nt V

aria

ble

crisi

s as

mea

sure

d by

sig

nific

ant n

egat

ive

retu

rns

in c

ount

ry B

at t

The

cons

ider

ed c

rises

per

iods

are

Mex

ico

1994

/5, T

haila

nd 1

997,

Rus

sia 1

998,

Bra

zil 1

999,

Tur

key

2001

, Arg

entin

a 20

02

(1)

(2)

(3)

(4)

(5)

(6)

pool

ed p

robi

t all

cris

es, e

mer

ging

m

arke

ts

linea

r pro

babi

lity

all c

rises

, em

ergi

ng

mar

kets

OLS

, all

cris

es,

emer

ging

mar

kets

pool

ed p

robi

t all

cris

es, e

mer

ging

m

arke

ts

linea

r pro

babi

lity

all c

rises

, em

ergi

ng

mar

kets

OLS

, all

cris

es,

emer

ging

mar

kets

0.10

9**

0.02

9**

0.00

3(0

.055

)(0

.012

)(0

.005

)0.

281*

**0.

074*

**-0

.018

**(0

.096

)(0

.020

)(0

.008

)2.

976

0.50

6-0

.147

3.38

40.

569

0.06

0(4

.459

)(0

.985

)(0

.392

)(4

.385

)(0

.953

)(0

.381

)5.

715

1.54

0-0

.510

3.39

71.

063

-0.5

09(5

.363

)(1

.135

)(0

.452

)(5

.343

)(1

.119

)(0

.447

)-0

.000

-0.0

000.

000*

-0.0

00-0

.000

0.00

0**

(0.0

00)

(0.0

00)

(0.0

00)

(0.0

00)

(0.0

00)

(0.0

00)

0.01

3**

0.00

2**

-0.0

000.

012*

*0.

002*

-0.0

00(0

.006

)(0

.001

)(0

.000

)(0

.006

)(0

.001

)(0

.000

)0.

194*

**0.

029*

**-0

.018

***

0.22

0***

0.03

3***

-0.0

23**

*(0

.056

)(0

.011

)(0

.004

)(0

.056

)(0

.010

)(0

.004

)5.

682

1.66

3-1

.511

***

8.52

42.

117

-1.7

08**

*(6

.159

)(1

.308

)(0

.521

)(6

.426

)(1

.308

)(0

.522

)-0

.039

-0.0

06-0

.002

-0.0

23-0

.003

-0.0

03(0

.058

)(0

.011

)(0

.004

)(0

.057

)(0

.010

)(0

.004

)-0

.082

*-0

.021

**0.

003

-0.0

84*

-0.0

21**

0.00

2(0

.046

)(0

.009

)(0

.004

)(0

.046

)(0

.009

)(0

.004

)-0

.044

-0.0

110.

003

-0.0

33-0

.009

0.00

2(0

.050

)(0

.011

)(0

.004

)(0

.051

)(0

.010

)(0

.004

)tim

e ef

fect

s ye

sye

sye

sye

sye

sye

sco

untr

y ef

fect

sye

sye

sye

sye

sye

sye

sC

onst

ant

-530

.683

***

-71.

981*

**12

.395

**-5

01.9

49**

*-6

5.41

0***

10.4

16**

(94.

537)

(12.

951)

(5.1

58)

(95.

941)

(13.

010)

(5.1

98)

Obs

erva

tions

539

571

571

539

571

571

R-s

quar

ed0.

280.

100.

290.

10St

anda

rd e

rror

s in

par

enth

eses

, * s

igni

fican

t at 1

0%; *

* si

gnifi

cant

at 5

%; *

** s

igni

fican

t at 1

%

lag

of fi

nanc

ial r

isk

inde

x

lag

of s

tock

mar

ket v

olat

iltiy

lag

of e

cono

mic

risk

inde

x

lag

of p

oliti

cal r

isk

inde

x

lag

of m

ean

grow

th e

xpec

tatio

n

lag

of in

tera

ct. o

vere

xp. i

ndex

&

cris

is in

A

lag

of in

tera

ct. u

ncer

t. in

B &

cr

isis

in a

ll in

itial

cris

is c

ntrs

.

lag

of in

tera

ct. c

om c

redi

tors

&

cris

is in

Ala

g of

inte

ract

. tra

desh

are

& c

risis

in

Ala

g of

inte

ract

. mar

ket s

ize

& c

risis

in

A

lag

of in

tera

ct. u

ncer

t. in

B &

cr

isis

in M

ex, R

us, T

ha

Table 3.10: Step 2: Effect of uncertainty in B on crisis there, additional controlfor common overexposed fund investors

3.6. APPENDIX 147

Pan

el r

egre

ssio

ns o

f c

risi

s in

pot

enti

ally

aff

ecte

d co

untr

y on

the

lagg

ed u

ncer

tain

ty in

BS

tep

2 E

ffect

of u

ncer

tain

ty in

a c

ount

ry o

n th

e pr

obab

ility

of a

cris

is th

ere

if a

cris

is ta

kes

plac

e in

an

initi

al c

risis

cou

ntry

Dep

ende

nt V

aria

ble

crisi

s as

mea

sure

d by

sig

nific

ant n

egat

ive

retu

rns

in c

ount

ry B

at t

Dat

a so

urce

: M

SC

IT

he c

onsi

dere

d cr

ises

per

iods

are

Mex

ico

1994

/5, T

haila

nd 1

997,

Rus

sia 1

998,

Bra

zil 1

999,

Tur

key

2001

, Arg

entin

a 20

02

(1)

(2)

(3)

(4)

(5)

(6)

pool

ed p

robi

t all

cris

es, a

ll co

untri

es

pool

ed p

robi

t all

cris

es, e

mer

ging

m

arke

ts

linea

r pro

babi

lity

all c

rises

, all

coun

trie

s

linea

r pr

obab

ility

al

l cris

es,

emer

ging

m

arke

ts

OLS

, all

crise

s,

all c

ount

ries

OLS

, all

cris

es,

emer

ging

m

arke

ts

0.29

2**

0.26

7*0.

061*

**0.

043*

*0.

005

0.01

1(0

.134

)(0

.140

)(0

.013

)(0

.019

)(0

.006

)(0

.008

)-0

.026

-0.0

38-0

.007

***

-0.0

08**

-0.0

04**

*-0

.003

**(0

.026

)(0

.027

)(0

.002

)(0

.004

)(0

.001

)(0

.002

)0.

318*

*0.

272*

0.01

8*0.

028

-0.0

08*

-0.0

08(0

.145

)(0

.159

)(0

.009

)(0

.018

)(0

.004

)(0

.008

)0.

608*

**0.

650*

**0.

057*

**0.

116*

**-0

.021

***

-0.0

26**

*(0

.157

)(0

.172

)(0

.011

)(0

.022

)(0

.005

)(0

.010

)0.

391*

0.55

5**

0.09

6***

0.19

1***

0.04

2***

0.03

7***

(0.2

02)

(0.2

23)

(0.0

15)

(0.0

29)

(0.0

07)

(0.0

13)

0.81

8***

0.86

9***

0.07

6***

0.13

1***

-0.0

32**

*-0

.055

***

(0.1

02)

(0.1

13)

(0.0

08)

(0.0

15)

(0.0

04)

(0.0

07)

-0.0

590.

026

0.00

40.

016

-0.0

02-0

.001

(0.1

30)

(0.1

41)

(0.0

08)

(0.0

16)

(0.0

04)

(0.0

07)

coun

try

effe

cts

yes

yes

yes

yes

yes

yes

Con

stan

t-2

.741

***

-1.7

14**

*-0

.032

-0.0

090.

028*

**0.

027*

(0.4

12)

(0.3

42)

(0.0

21)

(0.0

33)

(0.0

11)

(0.0

14)

Obs

erva

tions

2828

1630

4016

1947

4113

1946

R-s

quar

ed0.

150.

200.

030.

05St

anda

rd e

rror

s in

par

enth

eses

, * s

igni

fican

t at 1

0%; *

* si

gnifi

cant

at 5

%; *

** s

igni

fican

t at 1

%

retu

rns

as d

ep. V

aria

ble

lag

of c

risis

in R

us

lag

of c

risis

in T

ha

lag

of c

risis

in T

ur

lag

of u

ncer

tain

ty in

B

lag

of m

ean

of g

row

th

exp.

lag

of c

risis

in A

rg

lag

of c

risis

in B

ra

Table 3.11: Step 2: Effect of uncertainty in B on crisis there, robustness check

148 CHAPTER 3. UNCERTAINTY AND SPREAD OF CRISESM

argi

nal e

ffec

ts o

f pro

bit r

egre

ssio

ns o

f cr

isis

in p

oten

tial

ly a

ffec

ted

coun

try

on t

he la

gged

unc

erta

inty

Step

2 E

ffect

of u

ncer

tain

ty in

a c

ount

ry o

n th

e pr

obab

ility

of a

cris

is th

ere

if a

crisi

s ta

kes

plac

e in

an

initi

al c

risis

coun

try

Dep

ende

nt V

aria

ble

cris

is as

mea

sure

d by

sig

nific

ant n

egat

ive

retu

rns

in c

ount

ry B

at t

Dat

a so

urce

: M

SC

ITh

e co

nsid

ered

cris

es p

erio

ds a

re M

exic

o 19

94/5

, Tha

iland

199

7, R

ussi

a 19

98, B

razi

l 199

9, T

urke

y 20

01, A

rgen

tina

2002

(1

)(2

)(3

)(4

)(5

)(6

)(7

)(8

)

all c

ount

ries

emer

ging

m

arke

tsal

l cou

ntrie

sem

ergi

ng

mar

kets

all c

ount

ries

emer

ging

m

arke

tsal

l cou

ntrie

sem

ergi

ng

mar

kets

0.02

5***

0.03

9***

0.04

2**

0.05

2***

(0.0

07)

(0.0

11)

(0.0

17)

(0.0

19)

0.00

9**

0.01

6**

0.01

60.

020*

(0.0

04)

(0.0

07)

(0.0

10)

(0.0

10)

0.48

1*0.

637

0.51

1*0.

626

0.30

10.

629

0.25

50.

550

(0.2

63)

(0.4

31)

(0.2

74)

(0.4

45)

(0.7

73)

(0.8

18)

(0.7

85)

(0.8

27)

0.50

0**

0.72

8**

0.60

7***

0.90

7***

0.71

30.

631

1.06

41.

057

(0.2

24)

(0.3

26)

(0.2

29)

(0.3

34)

(0.9

56)

(0.9

88)

(0.9

55)

(0.9

86)

-0.0

00-0

.000

-0.0

00-0

.000

0.00

0-0

.000

0.00

0-0

.000

(0.0

00)

(0.0

00)

(0.0

00)

(0.0

00)

(0.0

00)

(0.0

00)

(0.0

00)

(0.0

00)

0.00

2**

0.00

2**

0.00

2**

0.00

2**

(0.0

01)

(0.0

01)

(0.0

01)

(0.0

01)

0.01

3***

0.01

7***

0.01

1***

0.01

5***

0.04

2***

0.04

1***

0.03

8***

0.03

6***

(0.0

03)

(0.0

05)

(0.0

04)

(0.0

06)

(0.0

10)

(0.0

10)

(0.0

10)

(0.0

10)

0.96

8**

1.10

7*0.

862*

*0.

922

1.83

81.

584

1.38

41.

051

(0.3

87)

(0.5

67)

(0.3

87)

(0.5

76)

(1.1

39)

(1.2

03)

(1.0

86)

(1.1

44)

-0.0

04-0

.008

-0.0

06-0

.010

-0.0

02-0

.004

-0.0

05-0

.007

(0.0

04)

(0.0

06)

(0.0

04)

(0.0

06)

(0.0

10)

(0.0

11)

(0.0

10)

(0.0

11)

-0.0

04-0

.006

-0.0

04-0

.006

-0.0

14*

-0.0

16*

-0.0

14*

-0.0

15*

(0.0

03)

(0.0

05)

(0.0

03)

(0.0

05)

(0.0

08)

(0.0

09)

(0.0

08)

(0.0

09)

-0.0

010.

001

-0.0

02-0

.000

-0.0

09-0

.006

-0.0

10-0

.008

(0.0

03)

(0.0

05)

(0.0

04)

(0.0

05)

(0.0

09)

(0.0

09)

(0.0

09)

(0.0

09)

time

effe

cts

yes

yes

yes

yes

yes

yes

yes

yes

coun

try

effe

cts

yes

yes

yes

yes

yes

yes

yes

yes

Obs

erva

tions

1309

903

1309

903

586

539

586

539

Stan

dard

err

ors

in p

aren

thes

es, *

sig

nific

ant a

t 10%

; **

sign

ifica

nt a

t 5%

; ***

sig

nific

ant a

t 1%

lag

of in

tera

ct. m

arke

t siz

e &

cr

isis

in A

lag

of in

tera

ct. o

vere

xp. i

ndex

&

cris

is in

A

lag

of m

ean

grow

th e

xpec

tatio

n

lag

of p

oliti

cal r

isk

inde

x

lag

of s

tock

mar

ket v

olat

ility

lag

of e

cono

mic

risk

inde

x

lag

of fi

nanc

ial r

isk

inde

x

lag

of in

tera

ct. u

ncer

t. in

B &

cr

isis

in M

ex, R

us, T

hala

g of

inte

ract

. unc

ert.

in B

&

cris

is in

all

initi

al c

risis

cnt

rs.

lag

of in

tera

ct. c

om c

redi

tors

&

cris

is in

Ala

g of

inte

ract

. tra

desh

are

&

cris

is in

A

Table 3.12: Marginal effect of uncertainty in B on probability of a crisis there

Bibliography

Acemoglu, D., S. Johnson, and J. Robinson (2001). The colonial origins of

comparative development: An empirical investigation. American Economic

Review 91 (5), 1369–1401.

Albuquerque, R., N. Loayza, and L. Serven (2005). World market integra-

tion through the lens of foreign direct investors. Journal of International

Economics 66 (2), 267–295.

Alfaro, L., S. Kalemli-Ozcan, and V. Volosovych (2005). Why doesn’t capital

flow from rich to poor countries: An empirical investigation. NBER Working

Paper, No. 11901 .

Arnold, I. and E. Vrugt (2006). Stock market volatility and macroeconomic

volatility. NRG Working Paper, No. 06-08 .

Bannier, C. (2006). The role of information disclosure and uncertainty in the

1994/95 mexican peso crisis: Empirical evidence. Review of International

Economics 14 (5), 883–909.

Bauer, C., B. Herz, and V. Karb (2007). Are twin currency and debt crises

something special? Journal of Financial Stability 3 (1), 59–84.

Berg, A. and C. Patillo (1998). Are currency crises predictable: A test. IMF

Working Paper, No. 154 .

Broner, F., G. Gelos, and C. Reinhart (2006). When in peril, retrench:

Testing the portfolio channel of contagion. Journal of International Eco-

nomics 69 (1), 203–230.

Broner, F. and R. Rigobon (2005). Why are capital flows so much more volatile

in emerging than in developed countries. Central Bank of Chile Working

Papers, No. 328 .

149

150 BIBLIOGRAPHY

Caballero, R. and K. Cowan (2006). Financial integration without the volatil-

ity. Central Bank of Chile Working Papers, No. 387 .

Caballero, R. and A. Krishnamurthy (2001). International and domestic collat-

eral constraints in a model of emerging market crises. Journal of Monetary

Economics 48 (3), 513–548.

Caballero, R. and S. Panageas (2005). A quantitative model of sudden stops

and external liquidity management. NBER Working Paper, No. 11293 .

Caballero, R. and S. Panageas (2006). Hedging sudden stops and precautionary

contractions. Journal of Development Economics (forthcoming).

Calomiris, C. and G. Gorton (1991). The origin of banking panics: Models,

facts and bank regulation. In R. Hubbard (Ed.), Financial Markets and

Financial Crises. Chicago: University of Chicago Press.

Calvo, G. (1998a). Capital flows and capital-market crises: The simple eco-

nomics of sudden stops. Journal of Applied Economics 1 (1), 35–54.

Calvo, G. (1998b). Growth, debt and economic transformation: The capital

flight problem. In F. Coricelli, F. Di Matteo, and F. Hahn (Eds.), New

Theories in Growth and Development. Palgrave Macmillan.

Calvo, G. (2003). Explaining sudden stop, growth collapse and bop crises -

the case of distortionary output taxes. IMF Staff Papers 50 (special issue).

Calvo, G., A. Izquierdo, and R. Loo-Kung (2006). Relative price volatility

under sudden stops: The relevance of balance sheet effects. Journal of

International Economics 69 (1), 231–254.

Calvo, G., A. Izquierdo, and L. Mejia (2004). On the empirics of sudden stops:

The relevance of balance-sheet effects. NBER Working Papers, No. 10520 .

Calvo, G., A. Izquierdo, and E. Talvi (2005). Sudden stop, financial factors,

and economic collapse in latin america: Learning from argentina and chile.

NBER Working Paper, No. 11153 .

Calvo, G., A. Izquierdo, and E. Talvi (2006). Sudden stops and phoenix

miracles in emerging markets. American Economic Review 96 (2), 405–410.

BIBLIOGRAPHY 151

Calvo, G., L. Leiderman, and C. Reinhart (1993). Capital inflows to latin

america: The role of external factors. IMF Staff Papers 40 (1), 108–151.

Calvo, G., L. Leiderman, and C. Reinhart (1996). Inflows of capital to devel-

oping countries in the 1990s. The Journal of Economic Perspectives 10 (2),

123–139.

Calvo, G. and E. Mendoza (2000). Rational contagion and the globalization

of securities markets. Journal of International Economics 51 (1), 79–113.

Calvo, G. and C. Reinhart (2000). When capital flows come to a sudden stop:

Consequences and policy. In P. Kenen and A. Swoboda (Eds.), Reforming

the International Monetary and Financial System. Washington DC: Inter-

national Monetary Fund.

Caprio, G. and D. Klingebiel (2002). Episodes of systemic and borderline

banking crises. In D. Klingebiel and L. Laeven (Eds.), Managing the Real and

Fiscal Effects of Banking Crises, World Bank Discussion Papers. Washington

DC: World Bank.

Caramazza, F., L. Ricci, and S. R. (2004). International financial contagion in

currency crises. Journal of International Money and Finance 23 (1), 51–70.

Carlsson, H. and E. van Damme (1993). Global games and equilibrium selec-

tion. Econometrica 61 (5), 989–1018.

Cavallo, E. and A. Frankel (2004). Does openness to trade make countries more

vulnerable to sudden stops, or less: Using gravity to establish causality. KSG

Working Paper, No. RWP04-038 .

Chari, V., P. Kehoe, and E. McGrattan (2005). Sudden stops and output

drops. American Economic Review 95 (2), 381–387.

Christiano, L., C. Gust, and J. Roldos (2004). Monetary policy in a financial

crisis. Journal of Economic Theory 119 (1), 64–103.

Chuhan, P., S. Claessen, and N. Mamingi (1987). Equity and bond flows to

latin america and asia: the role of global and country factors. World Bank

Policy Research Working Paper, No. 1160 .

152 BIBLIOGRAPHY

Corsetti, G., A. Dasgupta, S. Morris, and H. Shin (2004). Does one soros

make a difference: A theory of currency crises with large and small traders.

Review of Economic Studies 71 (1), 87–113.

Cukiermann, A. and A. Meltzer (1986). A theory of ambiguity, credibility and

inflation under discretion and asymmetric information. Econometrica 54 (5),

1099–1128.

Dasgupta, A. (2004). Financial contagion through capital connections: A

model of the origin and spread of banking panics. Journal of the European

Economic Association 2 (6), 1049–1084.

Diamond, D. and P. Dybvig (1983). Bank runs, deposit insurance, and liquid-

ity. Journal of Political Economy 91 (3), 401–419.

Diaz-Alejandro, C. (1983). Stories of the 1930s to 1980s. In A. Armella,

R. Dornbusch, and M. Obstfeld (Eds.), Financial Policies and the World

Capital Market: The Problem of Latin American Countries, pp. 5–35.

Chicago: University of Chicago Press.

Didier, T., P. Mauro, and S. Schmukler (2006). Vanishing contagion. IMF

Working Paper, No. 1 .

Disyatat, P. and G. Gelos (2001). The asset allocation of emerging market

mutual funds. IMF Working Paper, No. 111 .

Doenges, J. and F. Heinemann (2001). Competition for order flow as a coor-

dination game. Goethe Universitaet Frankfurt Working Paper, No. 64 .

Dornbusch, R., S. Claessens, and Y.-C. Park (1999). Contagion: How it spreads

and how it can be stopped. Mimeo, The World Bank .

Dornbusch, R., I. Goldfajn, and R. Valdes (1995). Currency crises and col-

lapses. Brookings Papers on Economic Activity, No. 2 , 219–293.

Drazen, A. (2000). Political contagion in currency crises. In P. Krugman (Ed.),

Currency Crises. Chicago: University of Chicago Press.

Durdu, C. and E. Mendoza (2006). Are asset price guarantees useful for

preventing sudden stops: A quantitative investigation of the globalization

hazard-moral hazard tradeoff. Journal of International Economics 69 (1),

84–119.

BIBLIOGRAPHY 153

Edwards, S. (1991). Capital flows, foreign direct investment and debt-equity

swaps in developing countries. In H. Siebert (Ed.), Capital Flows in the

World Economy, pp. 255–281. Tuebingen: Mohr.

Edwards, S. (2004a). Financial openness, sudden stops and current account

reversals. American Economic Review 94 (2), 59–64.

Edwards, S. (2004b). Thirty years of current account imbalances, current

acount reversals, and sudden stops. IMF Staff Papers 51 (special issue).

Edwards, S. (2005). Capital controls, sudden stops, and current account re-

versals. NBER Working Paper, No. 11170 .

Eichengreen, B., P. Gupta, and A. Mody (2006). Sudden stops and imf-

supported programs. NBER Working Paper, No. 12235 .

Eichengreen, B. and R. Hausmann (2005). Other People’s Money: Debt

Denomination and Financial Instability in Emerging Market Economies.

Chicago: University of Chicago Press.

Eichengreen, B., A. Rose, and C. Wyplosz (1996). Contagious currency crises:

First tests. Scandinavian Journal of Economics 98 (4), 463–84.

Fernandez-Arias, E. (1996). The new wave of private capital inflows: Push or

pull? Journal of Development Economics 48 (2), 389–418.

Flood, R. and P. Garber (1984). Collapsing exchange-rate regimes. Journal of

International Economics 17 (1), 1–13.

Forbes, K. and R. Rigobon (2001). Measuring contagion: Conceptual and em-

pirical issues. In S. Claessens and K. Forbes (Eds.), International Financial

Contagion. Kluwer Academic Publishers.

Frankel, J. and A. Rose (1996). Currency crashes in emerging markets: Em-

pirical indicators. CEPR Discussion Papers, No. 1349 .

Fratzscher, M. (2003). On currency crises and contagion. International Journal

of Finance and Economics 8 (2), 109–130.

Gelos, G. and S. Wei (2005). Transparency and international portfolio holdings.

The Journal of Finance 60 (6), 2987–3019.

154 BIBLIOGRAPHY

Gerlach, S. and F. Smets (1996). Contagious speculative attacks. CEPR

Discussion Paper, No. 1055 .

Giordani, P. and P. Soederlind (2003). Inflation forecast uncertainty. European

Economic Review 47 (6), 1037–1059.

Glick, R. and A. Rose (1999). Contagion and trade: why are currency crises

regional. Journal of International Money and Finance 18 (4), 603–617.

Goldstein, I. and A. Pauzner (2004). Contagion of self-fulfilling financial crises

due to diversification of investment portfolios. Journal of Economic The-

ory 119 (1), 151–183.

Goldstein, M. (1998). The Asian Financial Crisis. Washington DC: Institute

for International Economics.

Heinemann, F. (2005). Die theorie globaler spiele: Private information

als mittel zur vermeidung multipler gleichgewichte. Journal fuer Betrieb-

swirtschaft 55 (3), 209–241.

Heinemann, F. and G. Illing (2002). Speculative attacks: Unique equilibrium

and transparency. Journal of International Economics 58 (2), 429–450.

Hellwig, C. (2002). Public information, private information, and the multiplic-

ity of equilibria in coordination games. Journal of Economic Theory 107 (2),

191–222.

Hernandez, L. and R. Valdes (2001). What drives contagion: Trade, neighbor-

hood, or financial linkages. IMF Working Paper, No. 29 .

Hutchison, M. and I. Noy (2006). Sudden stops and the mexican wave: Cur-

rency crises, capital flow reversals and output loss in emerging markets.

Journal of Development Economics 79 (1), 225–248.

Kaminsky, G. and C. Reinhart (1999). The twin crises: The causes of banking

and balance of payments problems. American Economic Review 89 (3), 473–

500.

Kaminsky, G. and C. Reinhart (2000). On crises, contagion, and confusion.

Journal of International Economics 51 (1), 145–168.

BIBLIOGRAPHY 155

Kaminsky, G., C. Reinhart, and C. Vegh (2000). The unholy trinity of financial

contagion. The Journal of Economic Perspectives 17 (4), 51–74.

Kodres, L. and M. Pritsker (2002). A rational expectations model of financial

contagion. Journal of Finance 57 (2), 769–799.

Kraay, A., N. Loayza, L. Serven, and J. Ventura (2005). Country portfolios.

Journal of the European Economic Association 3 (4), 914–945.

Krugman, P. and M. Obstfeld (2003). International Economics: Theory and

Policy (6 ed.). Boston: Addison-Wesley.

Krugmann, P. (1979). A model of balance-of-payments crisis. Journal of

Money, Credit, and Banking 11 (3), 311–325.

Kyle, A. and W. Xiong (2001). Contagion as a wealth effect. Journal of

Finance 56 (4), 1401–1440.

Levchenko, A. and P. Mauro (2006). Do some forms of financial flows help

protect from sudden stops? IMF Working Paper, No. 202 .

Lopez-Mejia, A. (1999). Large capital flows: A survey of the causes, conse-

quences and policy responses. IMF Working Paper, No. 17 .

Lucas, R. (1990). Why doesnt capital flow from rich to poor countries? Amer-

ican Economic Review 80 (2), 92–96.

Marin, D. and M. Schnitzer (2006). When is fdi a capital flow. CEPR Discus-

sion Papers, No. 5755 .

Markowitz, H. (1959). Portfolio Selection: Efficient Diversification of Invest-

ments. New York: John Wiley and Sons.

Mendoza, E. (2006a). Lessons from the debt-deflation theory of sudden stops.

American Economic Review 96 (2), 411–416.

Mendoza, E. (2006b). Sudden stops in an equilibrium business cycle model

with credit constraints: A fisherian deflation of tobin’s q. NBER Working

Paper, No. 12564 .

Mendoza, E. and K. Smith (2006). Quantitative implications of a debt-

deflation theory of sudden stops and asset prices. Journal of International

Economics 70 (1), 82–114.

156 BIBLIOGRAPHY

Metz, C. (2002). Private and public information in self-fulfilling currency crises.

Journal of Economics 76 (1), 65–85.

Milesi-Ferretti, G. and A. Razin (1998). Sharp reductions in current account

deficits: An empirical analysis. European Economic Review 42 (3-5), 897–

908.

Milgrom, P. and J. Roberts (1990). Rationalizability, learning and equilibrium

in games with strategic complementarities. Econometrica 58 (6), 1255–1278.

Mishkin, F. and E. White (2003). Stock market bubbles: When does interven-

tion work? Milken Institute Review 5 (2), 44–52.

Mondria, J. (2006a). Evidence of financial contagion through attention reallo-

cation. Working Paper available at http://individual.utoronto.ca/jmondria/.

Mondria, J. (2006b). Financial contagion and attention allocation. Working

Paper available at http://individual.utoronto.ca/jmondria/.

Montiel, P. and C. Reinhart (1999). Do capital controls and macroeconomic

policies influence the volume and composition of capital flows: Evidence

from the 1990s. Journal of International Money and Finance 18 (4), 619–

635.

Morris, S. and H. Shin (1998). Unique equilibrium in a model of self-fulfiling

currency attacks. The American Economic Review 88 (3), 587–597.

Morris, S. and H. Shin (2003). Global games: Theory and applications (pro-

ceedings of the eighth world congress of the econometric society). In M. De-

watripont, L. Hansen, and S. Turnovsky (Eds.), Advances in Economics and

Econometrics. Cambridge: Cambridge University Press.

Morris, S. and H. Shin (2004). Coordination risk and the price of debt. Euro-

pean Economic Review 48 (1), 133–153.

Neymeyer, P. and F. Perri (2004). Business cycles in emerging economies.

NBER Working Paper, No. 10387 .

Obstfeld, M. (1994). The logic of currency crises. Banque de France, Cahiers

Economiques et Monetaires 43, 189–213.

BIBLIOGRAPHY 157

Pescatori, A. and A. Sy (2004). Debt crises and the development of interna-

tional capital markets. IMF Working Paper, No. 44 .

Prati, A. and M. Sbracia (2002). Currency crises and uncertainty about fun-

damentals. IMF Working Paper, No. 3 .

Reinhart, C. (2002). Default, currency crises and sovereign credit ratings. The

World Bank Economic Review 16 (2), 151–170.

Reinhart, C. and K. Rogoff (2004). Serial default and the ’paradox’ of rich to

poor capital flows. American Economic Review 94 (2), 53–58.

Rigobon, R. (2003). On the measurement of the international propagation

of shocks: Is the transmission stable? Journal of International Eco-

nomics 61 (2), 261–283.

Rose, A. (2004). Do we really know that the wto increases trade? The

American Economic Review 94 (1), 98–114.

Sachs, J. (1984). Theoretical issues in international borrowing. Princeton

Essays in International Finance, No. 54 .

Sachs, J., A. Tornell, and A. Velasco (1996). Financial crises in emerging mar-

kets: The lessons from 1995. Brookings Papers on Economic Activity 27 (1),

147–99.

Schinasi, G. and T. Smith (2000). Portfolio diversification, leverage, and fi-

nancial contagion. IMF Staff Papers 47 (2).

Schnitzer, M. (2000). Debt vs. foreign direct investment: the impact of

sovereign risk on the structure of international capital flows. Econom-

ica 69 (273), 41–67.

Taketa, K. (2004). Contagion of currency crises across unrelated countries.

IMES Discussion Paper, No. 2004-E-22 .

Tillmann, P. (2004). Disparate information and the probability of currency

crises: Empirical evidence. Economics Letters 84 (1), 61–68.

Van Rijckeghem, C. and B. Weder (2001). Sources of contagion: is it finance

or trade? Journal of International Economics 54 (2), 293–308.

Wooldridge, J. (2002). Econometric Analysis of Cross Section and Panel Data.

Cambridge, Massachusetts: The MIT Press.

Yuan, K. (2004). Asymmetric price movements and borrowing constraints: a

rational expectations equilibrium. Journal of Finance 60 (1), 379–411.

Eidesstattliche Versicherung

Ich versichere hiermit eidesstattlich, dass ich die vorliegende Arbeit selbstandig

und ohne fremde Hilfe verfasst habe. Die aus fremden Quellen direkt oder indi-

rekt ubernommenen Gedanken sowie mir gegebene Anregungen sind als solche

kenntlich gemacht. Die Arbeit wurde bisher keiner anderen Prufungsbehorde

vorgelegt und auch noch nicht veroffentlicht.

Washington, den 3. Oktober 2007

Fritzi Kohler-Geib

Curriculum Vitae

February 2008 PhD (Dr. oec. publ) in economicsLudwig-Maximilians-Universitat, Munich

October 2003 PhD Program in Economics,- February 2008 Munich Graduate School of Economics

at Ludwig-Maximilians-Universitat, Munich

March-June 2006 Visiting Ph.D student,February-July 2007 Universitat Pompeu Fabra, Barcelona

October 2002 Master (lic. oec. HSG) in economicsUniversitat St. Gallen, St. Gallen

CEMS-Master in international management(Community of European Management Schools)Universitat St. Gallen, St. Gallen

September- Exchange student,December 2000 University of Michigan Business School, Ann Arbor

October 1999- Exchange student,June 2000 Hautes Etudes Commerciales, HEC, Paris

July 1997 Abitur,Amos Comenius Gymnasium, Bonn

the effect of uncertainty on the occurrence and spread of

Documents